Neurocognitive subprocesses of working memory performance

RESEARCH ARTICLE

Neurocognitive subprocesses of working memory performance

Agatha Lenartowicz1 & Holly Truong1& Kristen D. Enriquez1 & Julia Webster1 & Jean-Baptiste Pochon1

&

Jesse Rissman1& Carrie E. Bearden1

& Sandra K. Loo1& Robert M. Bilder1

Accepted: 23 May 2021# The Author(s) 2021

AbstractWorking memory (WM) has been defined as the active maintenance and flexible updating of goal-relevant information in a formthat has limited capacity and resists interference. Complex measures of WM recruit multiple subprocesses, making it difficult toisolate specific contributions of putatively independent subsystems. The present study was designed to determine whetherneurophysiological indicators of proposed subprocesses of WM predict WM performance. We recruited 200 individuals definedby care-seeking status and measured neural responses using electroencephalography (EEG), while participants performed fourWM tasks. We extracted spectral and time-domain EEG features from each task to quantify each of the hypothesized WMsubprocesses: maintenance (storage of content), goal maintenance, and updating. We then used EEG measures of each subpro-cess as predictors of task performance to evaluate their contribution to WM. Significant predictors of WM capacity includedcontralateral delay activity and frontal theta, features typically associated with maintenance (storage of content) processes. Incontrast, significant predictors of reaction time and its variability included contingent negative variation and the P3b, featurestypically associated with goal maintenance and updating. Broadly, these results suggest two principal dimensions that contributeto WM performance, tonic processes during maintenance contributing to capacity, and phasic processes during stimulus pro-cessing that contribute to response speed and variability. The analyses additionally highlight that reliability of features acrosstasks was greater (and comparable to that of WM performance) for features associated with stimulus processing (P3b and alpha),than with maintenance (gamma, theta and cross-frequency coupling).

Keywords Working memory . EEG . Alpha . Theta . Gamma . P3 . Cross-frequency coupling . Maintenance . Updating . Goalmaintenance . RDoC

Introduction

Working memory (WM) was described decades ago as the cog-nitive ability to store information briefly online after it is nolonger available to our senses (Baddeley, 1986). A workgroupconvened by NIMH subsequently defined WM as “...the activemaintenance and flexible updating of goal/task relevant informa-tion (items, goals, strategies, etc.) in a form that has limitedcapacity and resists interference. These representations: mayinvolve flexible binding of representations; may be characterizedby the absence of external support for the internally maintainedrepresentations; and are frequently temporary, though this maybe due to ongoing interference” (NIMH, 2011). The importance

of WM in cognitive neuroscience research is underscored by itsextensive theoretical consideration in cognitive psychology(Baddeley, 2002; Cowan, 2008; Miller, 1956; Oberauer, Süß,Schulze, Wilhelm, & Wittmann, 2000), a multitude of validatedassays (Luck & Vogel, 1997; Nee, Wager, & Jonides, 2007;Vogel & Machizawa, 2004), and an association with putativebiological mechanisms (D'Esposito et al., 1998; Fuster &Alexander, 1971; Goldman-Rakic, 1995) that have suggestedclear links between cognitive and neural function. WM impair-ments may comprise the most clinically salient cognitive disor-ders, having been reported across nearly all neurological condi-tions, including traumatic brain injury (Alexander, Stuss, Picton,Shallice, & Gillingham, 2007; Stuss & Benson, 1984) and neu-ropsychiatric disorders, including anxiety (Brewin, Gregory,Lipton, & Burgess, 2010), depression (Joormann & Gotlib,2008), substance abuse (Bechara &Martin, 2004), attention def-icit hyperactivity disorder (ADHD) (Loo et al., 2007), schizo-phrenia (Gold et al., 2010), and bipolar disorder (Bearden,Hoffman, & Cannon, 2001).

* Agatha [email protected]

1 University of California Los Angeles, Los Angeles, CA, USA

https://doi.org/10.3758/s13415-021-00924-7

/ Published online: 21 June 2021

Cognitive, Affective, & Behavioral Neuroscience (2021) 21:1130–1152

http://crossmark.crossref.org/dialog/?doi=10.3758/s13415-021-00924-7&domain=pdf

http://orcid.org/0000-0001-8015-8251

mailto:[email protected]

Despite the salience of WM in the neurocognitive assess-ment of neuropsychiatric syndromes, it has remained a chal-lenge to understand how the putative neural mechanisms ofWM contribute to WM performance and its impairments.Namely, WM is a complex behavioral construct with multiplecomponent processes that can be impaired through differentfunctional neuroanatomic pathways, and each of these maycontribute uniquely to poor WM performance. For instance,the putative component function "maintenance” can involveshort-term storage associated with local cortical activities(Compte, Brunel, Goldman-Rakic, & Wang, 2000; Lisman& Idiart, 1995), long-term memory functions that may recruitcortico-hippocampal interactions (Axmacher et al., 2007),and/or maintenance of the behavioral goals guiding the behav-ior that may rely on the fronto-parietal network (O'Reilly &Frank, 2006). The extent to which these diverse processes areinvolved in WM remains a topic of debate and likely variesmarkedly across different tasks (Cowan, 2008; Nee& Jonides,2011). In addition, effective maintenance relies on effectiveencoding and its relation to contexts (updating); it requiresinteractions of maintenance processes with sensory and motorcircuit functions, and/or with more complex semantic func-tions that engage extensive cortico-cortical networks. Thus,impairments in WM performance may stem from such inter-actions rather than from maintenance processes per se. Forexample, recent studies of ADHD report that visual attentionprocesses during encoding predict neural responses duringWMmaintenance and other aspects of task performance, sug-gesting that attentional interactions during updating, ratherthan maintenance processes, may account for a portion ofWM performance impairments in this population(Lenartowicz et al., 2014; Lenartowicz, Mazaheri, Jensen, &Loo, 2018). In contrast, in schizophrenia, the contribution ofgoal maintenance processes to deficits ofWMhave beenwell-documented, hypothesized to result from dysfunction offrontoparietal connectivity or corticostriatopallidothalamiccircuits (Barch & Dowd, 2010). These observations highlighta need to understand the differential contributions of the func-tional anatomic bases of WM subprocess to clinically impor-tant disorders of brain and behavior.

The present study aimed to contribute to this understand-ing. Motivated by the NIMH Research Domain Criteria(RDoC) initiative (RDoC, Insel et al., 2010), we defined pu-tative neural circuits considered important for WM, identifiedEEG indicators theoretically relevant to these circuits, andassessed their associations with WM behavioral measures ina heterogenous sample of individuals ascertained to captureclinically relevant variability in WM function. We did this intwo steps. First, as noted above, we defined neural circuitsputatively underlying WM: local-cortical circuits associatedwith short-term memory storage (Compte et al., 2000;Lisman & Idiart, 1995); cortico-hippocampal circuits neces-sary for WM operations that exceed span and enable long-

term memory storage (Axmacher et al., 2007); and fronto-parietal corticocortical circuits associated with goalmaintenance and interference control (O'Reilly & Frank,2006). WM “updating” is another important aspect of WMthat we conceptualized as the phasic alteration of local circuitsvia subcortical-cortical circuits (D'Ardenne et al., 2012), oc-curring during stimulus processing and response selection. Toprovide neurophysiological indicators of each subprocess, weused electroencephalographic (EEG) metrics, previously as-sociated with the putative neural circuits underlying the pro-posed neurocognitive constructs, during four different WMtasks, and then used dimension-reduction methods to con-struct simpler neurophysiological indicators of each circuit.The EEG signals include frontal gamma-range power, contra-lateral delay activity (CDA) and N170 potentials for short-term storage; frontal theta-range power, and gamma cross-frequency coupling (CFC) measures for long-term storage;alpha-range CFC and contingent negative variation (CNV)event-related potential (ERP) for goal maintenance; P3bERP, and alpha-range power for updating. Some of theseEEGmeasures were a priori linked to specificWM constructs(e.g., the CNV has long been seen as an EEG indicator ofmaintenance operations, and CDA during the lateralizedchange detection task has been validated as an index of WMcapacity), but the interpretation of other EEG signals maydepend on the task context within which these are collected.All measures, and the rationale for their selection, are de-scribed in detail in the Methods section. To test the validityof these derived measures, we regressed WM performance onthe EEG circuit indicators. In sum, we examine the contribu-tion of proposedWM functional circuits to WM performance,and do so both across WM tasks and in a clinically heteroge-neous population.

Methods and Materials

Participants

We recruited 200 adults from the community to participate inthe project: “Multi-Level Assays of Working Memory andPsychopathology”—supported by the National Institute ofMental Health (NIMH) Research Domains Criteria (RDoC)Initiative (PI: Bilder, R01-MH101478). To overcome biasesassociated with traditional diagnostic inclusion/exclusioncriteria, the project studied adults who were care-seeking(CS, n = 150) and non-care-seeking (NCS, n = 50), and pri-oritized enrollment of patients with more severe symptoms toensure a broad range of psychopathology and high variance inmajor symptom dimensions. The CS group comprised indi-viduals who were seeking treatment for mental or emotionalproblems and responded to advertisements or were referred tothe UCLA Neuropsychiatric Behavioral Health Service. The

1131Cogn Affect Behav Neurosci (2021) 21:1130–1152

NCS group comprised individuals who responded to adver-tisements and had not sought behavioral health or substance-abuse services within 12 months preceding enrollment.

Additionally, participants satisfied the following inclusioncriteria: aged 21-40 years; completed at least 8 years of formaleducation; general mental status, hearing, motor coordination,and cooperation sufficient to complete procedures; languageproficiency in English; IQ estimate >70 (on WAIS-IVVocabulary and Matrix Reasoning); visual acuity 20:50 orbetter within each eye (right and left eyes tested separately);no medical or neurological illness or treatment expected tohave cognitive effects; no psychotropic or sedating drugswithin 24 hours of exam; no long-acting antipsychotics orelectroconvulsive therapy within the preceding 6 months; nosubstance abuse disorder diagnosis other than caffeine or nic-otine in preceding 6 months; negative urinalysis results forTHC, cocaine, amphetamine, opiates, and benzodiazepines;no contraindications to MRI exam (metal in the body, claus-trophobia, or for females, pregnancy).

All participants underwent comprehensive diagnostic andneurocognitive assessments, the results of which have beenpresented elsewhere (pertinent details are provided inSupplemental Materials). In this report, we consider only asubset that are most directly relevant to the EEG features ex-amined: overall disability as indexed by the World HealthOrganization Disability Affective Schedule (WHODAS 2.0),32-item summary score, as well as indicators of anxiety anddepression (Brief Psychiatric Rating Scale Anxiety andDepression [BRIEF] subscales; Patient-Reported OutcomesInformation System [PROMIS] anxiety and depression sub-scales), known to modulate several of the EEG indicatorsmeasured.

Tasks

Participants completed four different WM tasks during whichEEG was recorded: lateralized change detection (LCD),Sternberg spatial working memory (SWM), delayed face rec-ognition (DFR), and dot-pattern expectancy task (DPX)(Figure 1). The tasks were presented in one of two orders,counterbalanced across participants (SWM, LCD, DPX,DFR or DFR, DPX, LCD, SWM). For all tasks, stimuli werepresented on a Dell PC (Round Rock, TX; 17” monitors with4:3 aspect ratio) and responses were collected on a QWERTYkeyboard, controlled by E-Prime Software (v2.0; PsychologySoftware Tools, Pittsburg, PA).

We note that tasks involving overt verbal responses werenot included, because tasks with these response characteristicsare generally less suitable for and therefore not as well vali-dated as the selected tasks that demand primarily button-pressresponses and have been validated in functional MRI para-digms. We additionally aimed to minimize the use of verbaltask stimuli to avoid possible confounding of specific WM

processes with individual differences in semantic knowledgebetween participants (given that those with more robust se-mantic capacities would require lessWM tomaintain the samematerial in mind). A larger battery of tasks, including othercomplex WM tasks and verbal WM tasks was completed sep-arately from the EEG sessions. To ensure the tasks were con-sistent with current definitions of workingmemory constructs,the tasks were selected from published recommendations oftwo current initiatives: CNTRICS - Cognitive NeuroscienceTreatment Research to Improve Cognition in Schizophrenia(Barch, Berman, et al., 2009a; Barch, Carter, et al., 2009b);and RDoC – Research Domain Criteria (Bilder et al., 2019;Insel et al., 2010). Each of the EEG tasks are described below.Wewill first describe the tasks during which EEGwas record-ed and then describe further evidence that the EEG signalsextracted during these tasks are related to specific circuitmechanisms, as noted above: local cortical circuits (short-termstorage); corticohippocampal circuits (long-term storage);frontoparietal corticocortical circuits (goal maintenance);and corticostriatopallidothalamic (CSTP) circuits (goalmaintenance & updating).

It should be noted that there is not an isomorphic relationbetween individual tasks and putative circuit inferences be-cause the circuit inference depends on when during the taskthe EEG signal was collected and other signal characteristics,including frequency, amplitude, and spatial location(s). Forexample, in the Spatial Working Memory task described be-low, EEG signal collected during the encoding phase is as-sumed to be related to different circuit mechanisms than EEGactivity collecting during the maintenance period. Other spe-cific signals (e.g., event-related decreases in alpha power, orthe P3b event-related potential) also merit unique interpreta-tions and are discussed below in the section on EEG FeatureExtraction.

Lateralized Change Detection (LCD) The LCD task uniquelycaptures the contralateral delay activity (CDA), a feature de-rived from the EEG signal during WM maintenance, that hasbeen proposed as a correlate of WM capacity. The CDA alsois a putative indicator of local circuit integrity, albeit there areidentified moderating roles of both long range corticocorticaland CSPT circuits on capacity (Leonard et al., 2013; Vogel &Machizawa, 2004). In this task, in a given trial, participantsare first presented with a cue that instructs them to rememberthe color of either circles or rectangles, in a subsequent 200-ms visual array containing rectangles on one side of fixationand circles on the other side, with side-allocation randomlyvaried across trial. The array is followed by an 800-ms main-tenance period during which only a fixation marker is presenton screen. Finally, a target array is presented that is eitheridentical to the encoding array or differs via a color changeof one of the attended elements (unattended elements wereunchanged). Participants have 2,000 ms to indicate by button

1132 Cogn Affect Behav Neurosci (2021) 21:1130–1152

press whether the target is same or different than the encodingarray. For each participant, trials varied randomly in attendedshape (circles/rectangles) and whether the target matched/didnot match the encoding array, with equal proportions of eachtrial type per subject. Furthermore, a given trial varied in load,namely the number of elements to remember (1, 3, or 5). Atotal of 36 trials were presented per block and participantsperformed a total of 6 blocks (216 total), resulting in 72 trialsper load. Our primary behavioral outcome was maximumstorage capacity (Leonard et al., 2013) (KLCD), defined as K= n * (HR-FA)/(1-FA), where n is the number of to-be-remembered objects, HR is the hit rate, and FA is the falsealarm rate; the maximum value across all values of n was used.We also recorded median reaction time (RTLCD) and standarddeviation of RT (SDRTLCD) for correct trials.

Spatial Working Memory (SWM) The SWM task is a delayedmatch-to-sample variant of the classic “Sternberg” paradigmthat is an archetypal measure of WM “maintenance” thoughtto depend on frontoparietal corticocortical and CSPT circuits(Glahn et al., 2002). Trials begin with a fixation cross present-ed for 500-msec, followed by a 2-sec encoding display con-taining 1, 3, 5, or 7 yellow dots whose locations are to be

remembered. The number of dots is a manipulation of load,with greater load expected to engage more WM. The screenthen turns blank for a 3-sec maintenance interval. A singleprobe dot then appears and remains onscreen for a maximumof 3 sec. During this time, participants indicate with a buttonpress (left and right arrows) whether this probe stimulus is in alocation previously shown (match) or not (non-match). Ablank screen follows for a 1.5- to 2-sec intertrial interval.For each participant, trials vary randomly in load and whethera match or non-match probe is presented, with equal numberof trials per condition per subject. Participants perform a totalof 96 trials (24 per load), presented in 4 blocks of 24 trials(with equal number of trials per condition per block). Primarybehavior outcome variables are maximum capacity, as well asreaction time (RTSWM) and its standard deviation (SDRTSWM)for correct trials. Capacity was defined following Cowan as, K= n * (H-FA), where n = load, H = hit rate, FA = false-alarmrate. Maximum capacity (KSWM) was defined as the maxi-mum capacity across loads.

Delayed Face Recognition (DFR) The DFR involves presenta-tion of novel face stimuli in a SWM task delayed match-to-sample format; this task has been used to identify

a

b

c

d

+

1 s

+

2.5 s

+

3 s 1 s

+

.5 s

+

2 s

+

3 s

+

max 3 s 1.5-2 s

RememberCircle

Until Press ‘s’

+

.2 s

+

.8 s

+

2 s 1 s

+

1 s

A

+

2 s .5 s

X

+

1-2 s

BFig. 1 Study participants performed four working memory tasks duringwhich EEG signals were acquired: (a) delayed face recognition (DFR),(b) Sternberg spatial working memory (SWM), (c) lateralized changedetection (LCD), and (d) dot-pattern expectancy (DPX) task. Each taskincluded a target-stimulus encoding phase, a maintenance phase during

which the target disappeared off screen, and a probe phase during aresponse was recorded. During the DPX task target and probe were de-fined by stimulus pairing. Namely, participants were to respond if a stim-ulus labeled as X, followed a target labeled as “A”, but not following anyother stimulus (i.e., “B”). See text for additional details


corticohippocampal circuit activations recruited during condi-tions of highmemory load (Rissman, Gazzaley, &D'Esposito,2008). Each trial begins with a 2.5-sec encoding stimulus thatconsists of four images presented simultaneously, one in eachquadrant around the fixation cross. The four images includeeither 1 or 3 face stimuli, reflecting low or high load, inter-spersed with scrambled faces. Following a 3-sec delay duringwhich a fixation cross is presented, a probe face appears for 1sec, and subjects respond indicating whether or not it matchesany of the initial faces (using left and right arrow keys). Ascreen containing only a fixation cross then appears for a 1-secinterstimulus interval. Trials vary randomly in load, gender offace stimuli, and whether a match or nonmatch probe is pre-sented, with equal number of trials per condition. The trialorder was generated once, and the same sequence was usedfor all participants. Participants perform a total of 96 trials (48per load), presented in 3 blocks of 32 trials. Primary behavioroutcome variables are maximum capacity during load 3(KDFR), as well as reaction time (RTDFR) and its standarddeviation (SDRTDFR) for correct trials. Capacity is definedas for the SWM task.

Dot Probe Expectancy (DPX) The DPX is a continuous perfor-mance task, developed as a variant of the widely validatedAX-CPT task (Barch et al., 2004; Cohen, Barch, Carter, &Servan-Schreiber, 1999), and was selected to assess WM goalmaintenance. The task has undergone fMRI validation, dem-onstrating replicable frontoparietal corticocortical activations(Barch, Moore, Nee, Manoach, & Luck, 2012; Poppe et al.,2016). The DPX variant was developed as a clinical tool(Henderson et al., 2012), with the original letter stimuli ofthe AX-CPT task replaced by spatial dot stimuli, and the num-ber of trials and interstimulus interval duration optimized tominimize testing time while maintaining sensitivity. Trials inthis task comprise pairs of stimuli, context and target, that canbe one of four types denoted AX, AY, BX, and BY.Participants are to respond (button press) during the targetonly for AX trials, namely if the context is stimulus A andthe target is stimulus X. Any response for trial types AY, BX,or BY is an error. The task thus assesses the extent to whichthe context (i.e., A or B stimulus) is effectively maintained toguide response to the target (i.e., X or Y stimulus). A trialbegins with presentation of the context stimulus for 1 sec,followed by a 2-sec maintenance interval during which onlya fixation cross is on screen. The target then appears for 500ms. The interstimulus interval follows, with duration drawnrandomly from a square distribution (min = 1 sec, max = 2sec). Participants perform a total of 216 trials, presented in 6blocks. In each block, there are 26 AX trials, 4 AY and BXtrials and 2 BY trials, presented in a randomized order, for atotal of 156 AX trials, 24 AY/BX trials, and 12 BY trials. Thetrial order was generated once, and the same sequence wasused for all participants. Behavioral outcome measures

included d’DPX (d-prime), as well as reaction time (RTDPX)and its standard deviation (SDRTDPX) for correct AX trials.D-prime was defined as z(HR)-z(FA), where z(HR) is the ztransform of AX hit rate and z(FA) is the z transform of AXfalse alarm rate. This task thus was an outlier in its measure ofWM integrity. Rather than measuring capacity via load ma-nipulation, WM integrity was quantified through signal detec-tion sensitivity (detection of X), which is dependent on main-tenance of context (A).

EEG Recording

While participants performed each of the fourWM tasks, EEGrecordings were collected using BioSemi Active Two system,containing 64 silver chloride electrodes positioned in accor-dance with the 10/20 System. No filters were applied duringacquisition – the Active Two system acquires fully DCcoupled signals. Electrode impedances were brought below20KΩ before task recording. Electrical signals were recordedusing BioSemi hardware and ActiView recording software(BioSemi B.V., Netherlands). EEG was recorded at 1024 Hzwith no reference (as implemented in BioSemi’s zero-refframework). Electrode locations were recorded prior to theEEG session using a Zebris digitizer (Zebris MedicalGmbH, Germany).

Behavioral Dimension Analysis

The analysis included the 12 behavioral outcomemeasures forthe four tasks: median RT and its standard deviation (DPX,DFR, SWM, DPX), as well as WM capacity (DPX, DFR,SWM) and d’ (DPX). The reaction time (RT) measures arenot always included in studies ofWM but have been related togoal updating and interference control (Kessler & Oberauer,2015); specifically, RT tends to be slower when updatingdemands increase and faster when participants master taskdemands and effective costs of updating lessen. The measuresof WM capacity are widely used directly to indicate the ca-pacity of the relevant neural systems to represent and maintainthe task-relevant information (Barch et al., 2012). D-prime hasbeen used widely to index freedom from interference(Haatveit et al., 2010). Because our goal was to obtain perfor-mance metrics representative of WM processes independentof task, our first objective was to identify common underlyingdimensions across these 12 measures. To do so, we performeda principal components analysis. Model assumptions weretested using Kaiser-Meyer-Olkin Measure of SamplingAdequacy (>0.6) and Bartlett’s Test of Sphericity (p <0.0001). The number of factors was selected automaticallybased on eigenvalue >1. Factors were rotated using Promaxoblique rotation, selected based on the presence of factor cor-relations >|0.32|, (r = −0.41) (Tabachnick & Fidell, 2007).Factor scores derived from this analysis served as outcome


variables in the regression models below (e.g., EEGPredictors of Behavioral WM Outcomes).

EEG Preprocessing

The EEG signals collected during the WM tasks were proc-essed using custom MATLAB (The Mathworks, Inc.) scriptscalling functions from the EEGLAB (v.11.03.b) software(Delorme & Makeig, 2004). Continuous EEG data were highpass filtered (>1 Hz), inspected for noisy electrodes, whichwere excluded from further analysis. It is notable that the 1-Hz high pass was selected, because it improves the quality ofthe data and component decomposition (Winkler, Debener,Muller, & Tangermann, 2015). We acknowledge that high-pass filters >0.3 Hz can distort the ERP (Tanner, Morgan-Short, & Luck, 2015; Tanner, Norton, Morgan-Short, &Luck, 2016); however, as we were evaluating effects withstrong priors on topography, latency and condition effects,we felt the benefit to the signal-to-noise justified any potentialdistortions. The data were re-referenced to average reference.Within each subject, epochs of gross movements and muscleartifact were identified and removed if signal power in thatepoch exceed the 85th percentile for >60% of the channels.The cleaned data were decomposed into source signals byindependent component analysis (ICA) using the binica pro-gram in EEGLAB (extended infomax algorithm, stoppingweight change set to 1e-7, maximum learning steps set to1000, see S2 in Supplemental Materials for additional details)(Lee, Girolami, & Sejnowski, 1999). Each IC time course isthought to reflect a putative cortical source generator, associ-ated with a single topography across electrodes. IC timecourses were analyzed in lieu of channel data in all subsequentanalyses (unless specifically noted), segmented into epochstime-locked to the onset of the encoding stimulus, as de-scribed in the subsequent section. Except where stated, werestricted analysis to IC time courses with frontal and occipitaltopography. We excluded other ICs for two reasons. First,frontal and occipital topography ICs encompass the range oftopographies of the EEG features analyzed. Related to thispoint, frontal and occipital ICs were also identified in mostsubjects’ data (n = 153). In contrast other IC types were in-consistently present across subjects, which would reduce thesample size of the final analysis. Features of interest wereidentified and extracted from IC time courses of each subjectbased on a priori criteria as follows.

EEG Feature Extraction

The EEG data were used to extract features hypothesized,based on current literature, to measure elements of three corefunctional subprocesses contributing to WMmaintenance: lo-cal cortical storage in short-term memory, long-term storageprocesses presumed to rely on cortico-hippocampal

interactions, and goal maintenance/interference controlthought to involve fronto-parietal circuitry.We also quantifieda key putative supporting sub-process, namely WM updating,that complements the WM maintenance process.Additionally, we explored the contribution of vigilance-related alerting responses to WM performance. We note thatall features were identified at the outset of the analysis. Asnoted below, dimension reduction steps were performed withautomatic selection criteria, to eliminate experimenter bias. Asummary of all of the measures, and subsequent steps of di-mension reduction and final selection, is presented in Figure 2(also Table 4 in Results).

Short-Term Storage To quantify local cortical processingthought to underly storage in short-term memory, we comput-ed power in the gamma band (30-50 Hz), from ICs with fron-tal and occipital topography, during the maintenance intervalof each task. Gamma power cortical activity has been reliablyobserved during WM maintenance (Roberts, Hsieh, &Ranganath, 2013), used to decode WM content (Polania,Paulus, & Nitsche, 2012), and associated with WM capacity(Sauseng et al., 2009), putatively linking this feature with alocal storage function (Lisman & Buzsaki, 2008; Lisman &Idiart, 1995). For each frequency and time point in an interval,the power was divided by the mean power during baseline(−900 to −300 ms) and log-transformed (10log10) to decibel(dB) units. These values were averaged across all frequenciesand time points to produce a single value per subject per task(4 for frontal topography ICs and 4 for occipital topographyICs). In addition, for the LCD task data, we computed thecontralateral delay activity (CDA), a feature derived fromthe EEG signal during WM maintenance, that has been wellvalidated as a correlate of WM capacity (Leonard et al., 2013;Vogel & Machizawa, 2004). Finally, we explored the ampli-tude of the N170 event-related potential (ERP) componentduring DFR encoding, averaged across electrodes O1 andO2. We hypothesized that, because of the documented rela-tionship of N170 to face processing (Rossion & Corentin,2011), this feature may be directly related to local storage offace encoding stimuli in this task. Amplitudes in a 60 mswindow around the N170 peak were averaged across epochs,and the mean voltage in the baseline preceding the fixationcue (−200 ms to −150 ms) was subtracted. In sum, we obtain-ed a total of 10 putative indices of short-term WM storage.

Long-Term Storage To quantify long-term storage, associatedwith cortico-hippocampal interactions, we computed power inthe theta band (3-7Hz) from ICs with frontal topography, dur-ing the maintenance interval of each task. Theta power in-creases during WM maintenance in proportion to WM load(Jensen & Tesche, 2002; Mitchell, McNaughton, Flanagan, &Kirk, 2008; Onton, Delorme, & Makeig, 2005) and has beenproposed to play a significant role in memory encoding, in


particular with respect to cortico-hippocampal interactions(Lisman & Buzsaki, 2008; Lisman & Idiart, 1995).Furthermore, the strength of cross-frequency-coupling(CFC) between theta phase and gamma amplitude is a predic-tive neurophysiological marker of WM (Axmacher et al.,2010; Canolty et al., 2006). As above, we calculated frontalgamma power for each frequency and time point in the main-tenance interval, the power was divided by the mean powerduring baseline (−900 to −300 ms) and log-transformed(10log10) to decibel (dB) units. These values were averagedacross all frequencies and time points in the time-frequencywindow to produce a single value per subject per task. Tocalculate theta-gamma CFC, we adapted the methods ofDaume et al. (2017) to quantify phase-amplitude couplingusing the Kullback-Leibler distance-based modulation index(MI) introduced by Tort et al. (2010). In addition, to impose anormalization factor on CFC values, the MI values were con-verted to z-scores using a surrogate distribution following themethod of Canolty et al. (2006). Theta-gamma CFCMIs werecalculated at both frontal and posterior ICs. In sum, we ob-tained a total of 12 putative indices of long-term WM storage(frontal theta for each of 4 tasks, theta-gamma CFC for frontaland occipital ICs for each of 4 tasks).

Goal Maintenance To quantify goal maintenance and control,thought to engage fronto-parietal circuitry, we measured CFCbetween alpha range (8-12 Hz) phase and gamma amplitudeduring the maintenance phase of each task, across ICs withfrontal or occipital topography. The motivation for this metric

stems from animal and human research supporting an associ-ation between alpha range signals and engagement of the at-tention system (Foxe & Snyder, 2011; Jensen & Mazaheri,2010; Klimesch, 1997), as well as the analogous theoreticalframework suggesting that such alpha-range attentional sig-nals modulate gamma-range neuronal firing (Griffiths et al.,2019; Leszczynski, Fell, & Axmacher, 2015; Roux &Uhlhaas, 2014). Alpha-gamma CFC MI’s were calculated asdescribed for theta-gamma CFC (c.f. Long-Term Storage),replacing theta phase, with alpha phase. In addition, for theDPX task, we calculated the mean amplitude, at electrodeFCz, of the contingent negative variation (CNV) ERP, in the100 ms immediately preceding target onset. The CNV is aslow, negative potential that occurs when expecting a targetthat is contingent on a prior stimulus (i.e., as X is contingenton A in DPX). It is thought to include motor and cognitiveoperations necessary to prepare for the expected event, withnumerous cortical sources in frontal cortex, associated withcore goal maintenance functions (Fuster, 1985; Rosahl &Knight, 1995). In sum, we extracted 9 features (alpha-gammaCFC across 4 tasks, in frontal and occipital ICs, as well asCNV at FCz).

WM Updating In addition to core WM maintenance subpro-cesses, we sought to evaluate supporting functions that arearguably critical to effective WM performance that are eitherconceptualized as different from or complementary to main-tenance operations. This includes updating and alerting; wehave written elsewhere about the likelihood that these

DFRP3, n170 (n=198)Frontal IC features (n=177)Occipital IC features (n=195)

SWMP3, P2 (n=199)Frontal IC features (n=173)Occipital IC features (n=196)

LCDP3, CDA (n=195)Frontal IC features (n=171)Occipital IC features (n=189)

DPXP3, CNV (n=194)Frontal IC features (n=168)Occipital IC features (n=188)

Recruited Sample n=200

Within-Feature Reliability Analysis

P3n170 fCFC fθ oCFC oα γ P3P2 fCFC fθ oCFC oα γ P3CDA fCFC fθ oCFC oα γ P3CNV fCFC fθ oCFC oα γ

Within-Feature Principal Component Analysis

P3 PC fθ PC oα PC

Multiple Regression of WM Capacity and RT/RTSD on EEG Features n=153

Fig. 2 Data analysis overview. Each participant performed four EEGtasks, fromwhich we extracted several features—associated with putativeWM subprocesses (Table 4). Frontal-topography independent component(IC) features were not available in all datasets, contributing to sample sizereduction. Features that were sampled across multiple tasks (P3b, fCFC,fθ, oCFC, oα, γ) were tested for reliability and then reduced via principalcomponent analysis to a single dimension or excluded from subsequentsteps if threshold was not met (Cronbach’s alpha ≤ 0.5). Features that

were unique to a given task (N170, P2, CDA, CNV) were automaticallyincluded in the final analysis. The final multiple regression was per-formed for datasets that contained all features (n = 153). A separateanalysis was performed for WM capacity and RT/RTSD. fCFC = frontalcross-frequency coupling; fθ = frontal theta; oCFC = occipital cross-frequency coupling; oα = occipital alpha; γ = gamma; CDA = contralat-eral delay activity; CNV = contingent negative variation


functions use common phasic “arousal” mechanisms to intro-duce flexibility into stimulus processing and response selec-tion (Bilder, 2012). Updating refers to subprocesses that occurduring stimulus encoding, when information entering via thevisual system must interact with the WM system to introducenew material for storage or to select a stimulus- and task-relevant response. We identified two EEG features to quantifythis function, alpha signal power decreases and P3b ERP am-plitude during encoding and probe stimuli. As noted above,alpha-range oscillatory signals are strongly implicated in thegating function of the attention system, that controls whichstimuli are processed and which stimuli are ignored or gatedout from further processing (Foxe & Snyder, 2011; Jensen &Mazaheri, 2010; Klimesch, 1997). During stimulus process-ing, event-related decreases (ERDs) in alpha power strengthenwith degree of processing (e.g., encoding with load, visualdetection) and accuracy of subsequent behavioral response(e.g., recall accuracy, detection accuracy) (Hanslmayr et al.,2005; Romei et al., 2008; Samaha & Postle, 2015), suggestingthat alpha ERDs play an important role in WM updating(Klimesch, 1997, 2012). Alpha ERD during encoding andprobe stimuli in each task were therefore calculated, followingprotocols as described for gamma (c.f., Short-Term Storage)and theta (c.f., Long-Term Storage), for ICs with occipitaltopography. Note that because we found alpha ERD was sig-nificantly correlated across encoding and probe events in alltasks (rDPX = 0.80, rSCAP = 0.72, rDFR = 0.56, rLCD = 0.80, p <0.0001), we additionally collapsed alpha ERD across theseevents. Finally, although alpha is typically defined as 8-12Hz, we have previously found that the full span of the effectextends upwards through approximately 16 Hz (Lenartowiczet al., 2014; Lenartowicz et al., 2019). Hence, we defined thealpha band as 12-16 Hz. This was validated by confirming that8-12 Hz and 12-16 Hz ERD values are significantly correlatedduring both encoding (rDPX = 0.74, rSCAP = 0.70, rDFR = 0.66,rLCD = 0.60, p < 0.0001) and probe (rDPX = 0.70, rSCAP = 0.73,rDFR = 0.68, rLCD = 0.65, p < 0.0001) events.

The second feature related to WM updating that we iden-tified was amplitude of the P3b ERP. Maximal over parietalscalp around 300-600 ms following stimulus onset, this posi-tive potential has been associated with the function ofupdating context in WM (Polich, 2007) and, in alternate for-mulations, with categorization of content within WM (Kok,2001; Nieuwenhuis, Aston-Jones, & Cohen, 2005). The P3bthus is directly related to encoding processes that interact withWM. To quantify the P3b, we identified the P3b peak in the300-600 ms window during each encoding and probe stimu-lus, and calculated the mean amplitude in the 50-ms windowaround the peak at electrode Pz, selected based on knownpriors (Lenartowicz, Escobedo-Quiroz, & Cohen, 2010). Asin the case of alpha ERD, P3b amplitude was significantlycorrelated across encoding and probe events in all tasks(rDPX = 0.81, rSCAP = 0.55, rDFR = 0.78, rLCD = 0.74, p <

0.0001), hence we collapsed P3b amplitude values acrossthese events.

Finally, we quantified alerting, as an exploratory index ofgeneral vigilance processes, reasoning that if participants aredisengaged from the task, their WM maintenance and conse-quently performance also may be compromised. We quanti-fied alerting via P2 ERP amplitude, in response to a fixationstimulus, present at the onset of each trial in the SWM task. Inprior studies, the P2 has been shown to increase for cues ofincreasing complexity (Lenartowicz et al., 2010), stimuli ofincreasing task relevance (Potts, 2004) and during fixation forindividuals with higher WM performance (Lenartowicz et al.,2014), suggesting that it may be a relevant indicator of analerting response at the onset of behaviorally important events.P2 amplitude was calculated at electrode Fz as the averagevoltage in the 50-ms window surrounding the peak in the150-250 window post-fixation. Thus, we obtained 9 addition-al EEG measures, 8 of WM updating (alpha ERD in each of 4tasks, P3b in each of 4 tasks), and also P2 during fixation inthe SWM task.

EEG Dimension Reduction

In sum, we extracted a total of 40 EEG features, some ofwhich were repeated across tasks (e.g., gamma or theta duringmaintenance) and others that were unique (e.g., CDA in LCDtask or CNV in DPX task). In order to reduce the number ofvariables, we conducted additional principal component anal-yses for each EEG feature that was present in all four tasks(Figure 2). This included gamma (c.f., Short-Term Storage),theta and theta-gamma CFC (c.f., Long-Term Storage), alpha-gamma CFC (c.f., Goal Maintenance & Control), as well asalpha ERD and P3b (c.f., WM Updating & Alerting). As forbehavioral dimension reduction, model assumptions weretested using Kaiser-Meyer-Olkin (KMO) Measure ofSampling Adequacy (>0.6) and Bartlett’s Test of Sphericity(p < 0.0001). The number of factors was selected automatical-ly based on eigenvalue >1, and only one PCA was run percluster of measures, eliminating experimenter bias at thisstage. Factors were rotated using Promax oblique rotation, toallow for the presence of factor intercorrelations. This analysiswas critical in testing whether the proposed subprocesses ofWM are common across multiple tasks, in which case wewould expect a single principal component for a given metricto load on all four tasks. Finally, to further test the validity oftheWM subprocesses, we also performed a reliability analysisacross the four tasks, for each of the above metrics, eliminat-ing from further analysis any indicators with Cronbach’salpha < 0.5. This is a liberal threshold, without prior expecta-tion on what a sufficient reliability threshold might be forthese metrics and chosen with the goal of eliminating highlyunreliable features. The final set of predictors produced by thisprotocol is summarized in Table 4 and Figure 2.


EEG Predictors of Behavioral WM Outcomes

The core question driving this study was to determinewhich sub-processes of WM, as quantified by selectEEG features, predict performance in WM. To addressthis question, we performed a confirmatory multiple re-gression analysis (SPSS v.25, IBM Corporation, Somers,NY). A separate analysis was conducted for each out-come variable, namely, component scores derived fromthe behavioral dimension PCA. As predictors, we in-cluded all EEG measures extracted for each of the pu-tative WM sub-processes described above. Covariate re-gressors included age, gender and sampling group (careseeking versus non-care seeking), included to accountfor confounding effects. Individual coefficients wereevaluated using t-tests. The final sample size in thisanalysis is 153. As shown in Figure 2, data loss result-ed from (a) initial loss due to issues with EEG datacollection, and (b) exclusion of individuals who didnot show both frontal and occipital ICs in the decom-position, meaning that all predictors could not be repre-sented in the model. These were treated as missing dataand excluded in the multiple regression. Finally, regres-sion coefficients at p < 0.05 are interpreted as signifi-cant. We do not apply multiple comparison correction(to account for two models, one for each performancedimension) as they address, in principle, independentbehavioral dimensions. Further, no corrections are war-ranted at the dimension reduction steps, as no inferencewas performed.

Load and Group Effects

To further support interpretation, we performed a repeated-measures ANOVA for each measure of interest within eachtask, with independent variables including the within-subject variable LOAD (low, high, varying by task), aswell as the between-subject variable sampling GROUP(care-seeking vs. non-care seeking). A significant LOADmain effect was important to validate the role of a givenEEG feature in WM capacity, because capacity is definedby amount of content stored. Note that LOAD was notmanipulated in the DPX task and so it was replaced bythe comparison of TRIAL type (A vs. B cues) in this anal-ysis. The GROUP variable was included as a covariate tocapture differences in EEG features as a function of groupdifferences in psychopathology. The GROUP variable val-idates the sampling method in this project, designed tocapture a broad range of psychopathology in the CS group.Finally, we report Cohen’s f effect sizes, recommended foranalyses involving F tests or ANOVA models (J. Cohen,1988). Small, medium and large f values are traditionallydefined as 0.02, 0.15, and 0.35, respectively.

Results

Sample Demographics

Sample demographics are shown in Table 1, task perfor-mance in Table 2. Following the sampling strategy, groupmembership is defined by Care-Seeking (CS) and Non-Care-Seeking (NCS) definitions rather than DSM-5 diag-noses, but the diagnoses observed in each group are shownin Table 1. The samples were comparable in age, genderdistribution, years of education, estimated IQ, annualhousehold income, and racial distribution. The samplingstrategy was validated by increased overall disabilityscores and increased presence of positive neuropsychiatricdiagnoses in the care-seeking (CS) participants relative tothe non-care-seeking group (NCS). The CS group alsoshowed lower capacity scores, slower RT and greater RTvariability across several, although not all, tasks (Table 2),a pattern that was replicated in the PCA-reduced dimensionscores as well (c.f., Behavioral Dimensions below).

Task Reliability Given some variability in group differences incapacity across tasks, we sought to also evaluate the reliabilityof the tasks. We first calculated Cronbach’s alpha for each ofaccuracy, RT, and RTSD, across the four tasks. Reliabilitywas good for all metrics (n = 195): Accuracy Cronbach’salpha = 0.78; RT Cronbach’s alpha = 0.72; RTSDCronbach’s alpha = 0.69. Pair-wise task correlations weresignificant across all pairs, and in the range of 0.35-0.62 foraccuracy, 0.31-0.58 for RT, and 0.30-0.51 for RTSD.

Behavioral Dimensions

The behavioral-measure PCA identified two componentswith eigenvalues >1, explaining 52% of total variance.Model assumptions were met (KMO = 0.73, Bartlett’s testof Sphericity, x2 = 1012, p < 0.0001). The rotated com-ponent loadings are shown in Table 3, with loadings >0.3considered as making a significant contribution to thecomponent. The first component explained 37.4% of totalvariance. Loadings on this component were greatest fromRT-related metrics, namely RT and SDRT for all fourtasks. A higher score on this component would indicateslower and more variable responding. The second compo-nent explained 14.5% total variance. Strongest loadingson this component corresponded to capacity indices (K)across tasks. Higher scores on this component would in-dicate higher capacity across tasks. Thus, the analysisidentified capacity and RT/RTSD as separable, albeit cor-related (r = −0.41), dimensions across four WM tasks.Individual scores on these components were extractedvia regression and used as WM outcomes below, withcomponent 1 scores referred to as a composite RT/


SDRT outcome variable and component 2 scores referredto as a composite capacity (K) outcome variable. A vali-dation t-test analysis indicated that composite RT/SDRTwas faster/lower (t(193) = 3.1, p = 0.002, M = −0.33 vs.0.14), and composite capacity was higher (t(193) = 2.5, p= 0.01, M = 0.27 vs. −0.11), in NCS individuals than CSindividuals.

Working Memory EEG Features

The final EEG features identified for eachWM subprocess aresummarized in Table 4 and illustrated in Figures 3, 4 and 5. A

total of 7 EEG features of the original 40 were retained aspredictors for subsequent analyses.

Short-Term Storage We retained only CDA (LCD task) andN170 (DFR task) indicators of short-term storage. Across-taskCronbach’s alpha for frontal maintenance gamma was 0.50and for occipital maintenance gamma was 0.46. Thus, gammashowed relatively poor to moderate reliability across tasks,and was not retained for further analysis (see Table S1 forPCA results for this measure).

Long-Term Storage A PCA on frontal theta during mainte-nance across tasks identified a single component, explaining

Table 1 Demographics

CS NCS Group difference

Sample size 142 58 -

% Females 66.2% 56.9% x2(1,200) = 1.5, p = 0.22

Mean age (SD) 28.4 (5.5) 27.1 (4.8) t(198) = 1.6, p = 0.11

Years of education (SD) 17.3 (1.8) 17.2 (1.8) t(198) < 1, p = 0.70

Estimated IQ (SD) 102.3 (11.6) 105.0 (12.4) t(194) = 1.5, p = 0.30

WHODAS 32-item Disability Score (SD) 20.6 (15.9) 6.12 (8.8) t = 6.5, p < 0.0001

Annual household income

<$10,000 9% 3% x2(5,200) = 2.9, p = 0.72$10,000-$19,999 8% 2%

$20,000-$39,999 15% 12%

$40,000-$59,999 17% 14%

$60,000-$99,999 15% 14%

>$99,999 20% 17%

Unreported 16% 38% X2(1,200) = 11.2, p = 0.001

Racial distribution

American Indian/Alaska Native 0.7% 1.7% X2 (7,200) = 1.7, = 0.95Asian 21.2% 24.1%

Black or African American 9.2% 8.6%

Hawaiian or Pacific Islander 2.8% 1.7%

More than one race 9.2% 5.2%

Unknown or unreported 11.3% 12.1%

White 45.8% 46.6%

Hispanic or Latino 31% 22.4% x2(1,200) = 3.0, p = 0.23

Not Hispanic or Latino 66.9% 77.%

Primary Diagnosis

Mood 53.5% 19.0% x2(6,200) = 59.8, p < 0.0001Anxiety 22.5% 10.3%

Psychosis 5.6% 0.0%

Eating disorder 0.7% 0.0%

Anti-social personality disorder 0.7% 0.0%

Attention deficit hyperactivity disorder 3.5% 3.4%

No diagnosis 13.4% 67.2%

SD = standard deviation; WHODAS = World Health Organization Disability Assessment Schedule 2.0; Significant differences (p < 0.05) indicated inbold


45.5% of total variance (KMO = 0.49, Bartlett’s test ofSphericity, x2 = 336, p < 0.0001, see Table S2 for componentloadings). Cronbach’s alpha for theta across tasks was 0.52,thus it was retained for further analysis. In contrast,Cronbach’s alpha for theta-gamma CFC measures was 0.06for parietal ICs and 0.02 for frontal ICs. Thus, theta-gammaCFCmeasures were excluded from further analysis due to lowreliability across tasks.

Goal Maintenance The only EEG feature included was CNV,measured in the DPX task. This was because for alpha-gammaCFC, Cronbach’s alpha was 0.02 and 0.13 for frontal and

posterior ICs respectively. Thus, alpha-gamma CFCmeasureswere excluded from further analysis.

WMUpdatingWith respect to WM updating, both alpha ERDand P3b were reliable, with Cronbach’s alpha across tasksexceeding 0.7 (alpha ERD = 0.72, P3b = 0.72). Also, bothmeasures resulted in a single component following PCA. Foralpha ERD, a single component explained 62.5% of total var-iance (KMO = 0.75, Bartlett’s test of Sphericity, x2 = 238, p <0.0001, see Table S4 for component loadings). For P3b, asingle component explained 63.3% of total variance (KMO= 0.76, Bartlett’s test of Sphericity, x2 = 253, p < 0.0001,

Table 2 Task performance

CS Mean (SD) NCS Mean (SD) Group difference

Delayed Face Recognition (DFR) Task Performance (n = 199)

KDFR 1.56 (0.53) 1.61 (0.42) t < 1, p = 0.5

RTDFR 0.94 (0.17) 0.90 (0.13) t = 1.85, p = 0.07

SDRTDFR 0.25 (0.06) 0.24 (0.06) t = 1.68, p = 0.09

AccuracyDFR 0.84 (0.09) 0.86 (0.05) t = 1.7, p = 0.09

#trials load1 41.4 (4.9) 41.3 (4.9) t < 1, p = 0.93

#trials load3 32.8 (5.3) 33.1 (5.4) t < 1, p = 0.72

Sternberg Spatial Working Memory (SWM) Task Performance (n=199)

KSWM 5.19 (1.13) 5.48 (1.13) t = 1.44, p = 0.15

RTSWM 1.10 (.22) 0.96 (0.20) t = 2.88, p = 0.004

SDRTSWM 0.34 (0.09) 0.32 (0.08) t = 2.05, p = 0.04

AccuracySWM 0.89 (0.08) 0.92 (0.05) t = 2.6, p = 0.01

#trials load1 20.6 (2.7) 20.8 (2.3) t < 1, p = 0.57

#trials load3 19.5 (3.1) 20.3 (2.4) t = 1.6, p = 0.11

#trials load5 18.6 (3.2) 19.0 (3.0) t < 1, p = 0.45

#trials load7 18.4 (3.2) 19.0 (2.7) t = 1.4, p = 0.19

Lateralized Change Detection Task Performance (n = 196)

KLCD (SD) 2.53 (0.89) 2.81 (0.70) t = 2.16, p = 0.03

RTLCD (SD) 0.60 (0.15) 0.55 (0.10) t = 2.27, p = 0.03

SDRTLCD (SD) 0.23 (0.08) 0.19 (0.06) t = 3.00, p = 0.003

AccuracyLCD (SD) 0.82 (0.09) .86 (.05) t = 2.8, p = 0.005

trials load1 26.4 (2.7) 26.7 (2.4) t < 1, p = 0.38

#trials load3 23.6 (3.5) 24.7 (3.0) t = 2, p = 0.05

#trials load5 20.2 (3.3) 20.9 (2.8) t = 1.4, p = 0.15

Dot-Pattern Expectancy Task Performance (n = 200)

d'DPX (SD) 3.00 (1.17) 3.37 (1.01) t = 2.13, p = 0.03

RTDPX (SD) 0.44 (0.08) 0.43 (0.04) t = 1.51, p = 0.13

SDRTDPX (SD) 0.12 (0.04) 0.11 (0.04) t = 2.52, p = 0.01

AccuracyDPX (SD) 0.94 (0.09) 0.96 (0.07) t = 1.6, p = 0.11

#trials AX 134.8 (21.3) 136.1 (11.3) t < 1, p = 0.67

#trials AY 20.9 (3.6) 20.9 (2.3) t < 1, p = 0.97

#trials BX 21.8 (2.0) 21.6 (2.2) t < 1, p = 0.58

#trials BY 10.6 (1.4) 10.6 (1.3) t < 1, p = 0.93

SD = standard deviation; K = capacity score as defined in text; RT = median reaction time (seconds); SDRT = standard deviation of reaction time(seconds), d’ = d-prime, #trials = number of accurate trials in given condition. Significant differences (p < 0.05) indicated in bold


see Table S3 for component loadings). Finally, alerting wasmeasured using P2 amplitude during fixation in SWM only;thus, no further reduction was needed.

EEG Predictors of WM Capacity The results of regressing com-posite capacity (K) on EEG features, across WM subpro-cesses, is shown in Table 5. Higher composite capacity scoreswere associated with two EEG features, both during the

maintenance interval: (1) more negative (and therefore stron-ger) CDA in the LCD task (p = 0.001), and (2) higher com-posite frontal theta scores (p = 0.008). Thus, composite capac-ity scores were most strongly associated with EEG indicatorsof short and long-term storage functions. At a more liberalfalse positive threshold of p < 0.1, higher composite capacityscores we also associated with higher P3b amplitude (p =0.06), an index of WM updating, and stronger CNV in DPXtask (p = 0.098), associated with goal maintenance.

Load Effects

To further aid interpretation, we also evaluated effects of loadon each of the significant variables, hypothesizing that featuresthat predict WM capacity should also be sensitive to manipula-tions of load. These analyses were performed on non-reduced(pre-PCA) data. Of the three variables that significantly predict-ed composite WM capacity, we found significant load effectsfor CDA and inconsistent effects for maintenance theta.Namely, the main effect of load was significant for CDA(F(2,193) = 14.8, p < 0.0001, Cohen’s f = 0.28). As expected,magnitude of CDA increased with load (load 1:M = −0.16 uV,SE = 0.044; load 3: M = −0.48 uV, SE = 0.049; load 5: M =−0.40 uV, SE = 0.048). Frontal theta showed main effects ofload in only two of four tasks, in the SWM task (F(3,171) = 2.6,p = 0.05,Cohen’s f = 0.13) and the LCD task (F(2,169) = 6.4, p= 0.002, Cohen’s f = 0.20). These effects were small in magni-tude, and inconsistent in direction, with theta power lowest inlow-load condition in the SWM task (M low to high: −0.37,

Table 3 Performance rotated component loadings

Measure RT/RTSD compositecomponent

Capacity compositecomponent

KDFR −0.025 0.393

KSWM 0.268 0.828

KLCD 0.063 0.705

d'DPX −0.025 0.763

RTDFR 0.737 0.175

RTSWM 0.699 −0.184RTLCD 0.934 0.329

RTDPX 0.527 −0.308SDRTDFR 0.644 −0.009SDRTSWM 0.416 −0.303SDRTLCD 0.767 −0.049SDRTDPX 0.411 −0.576

Loadings > 0.35 are in bold. Loadings > 0.3 are italicized to emphasizeclustering of effects. Note that higher RT/RTSD composite scores indi-cate slower/more variable response time. Higher capacity compositescores indicate better capacity

Table 4 Working memory EEG features

Construct Starting features Tasks included PCA Cronbach’s α acrosstasks

Variable included in finalanalysis

Short Term Storage Maintenance Gamma (frontal)* DFR SWM LCD DPX ✓ 0.46

Maintenance Gamma (posterior)* DFR SWM LCD DPX 0.50

CDA LCD - CDA amplitude

N170 DFR - N170 amplitude (O1 & O2)

Long Term Storage Maintenance Theta (frontal) DFR SWM LCD DPX ✓ 0.52 PC1: composite frontal theta

Maintenance Theta-Gamma CFC(frontal)*

DFR SWM LCD DPX 0.02 -

Maintenance Theta-Gamma CFC(posterior)*


Goal Maintenance &Control

Maintenance Alpha-Gamma CFC(frontal)*


Maintenance Alpha-Gamma CFC(posterior)*


CNV DPX - CNV amplitude (FCz)

WM Updating Stimulus P3b (Pz) DFR SWM LCD DPX ✓ 0.72 PC1: composite posteriorP3b

Stimulus Alpha ERD (posterior) DFR SWM LCD DPX ✓ 0.72 PC1: composite posterioralpha ERD

Alerting P2 (frontal) SWM - P2 amplitude (Fz)

PCA = principal component analysis; PC = principal component (scores); *Excluded from final analysis due to low Cronbach’s Alpha across tasks


−0.03, −0.19, −0.08 dB), but highest in low-load condition inthe LCD task (M low to high: −0.14, −0.34, −0.35 dB). Thusload-effect analyses provide convergent evidence for the role ofCDA, and inconsistent evidence for theta.

Group Effects There was no difference in CDA between CSand NCS participants (F(1,193) = 0.61, p = 0.43, Cohen’s f =0.05; CS:M = −0.32 uV, SE = 0.033; NCS:M = −0.37 uV, SE= 0.05). Similarly, groups did not differ in frontal theta: DFR(F(1,175) = 0.04, p = 0.85, Cohen’s f = 0.01), DPX (F(1,166)= 0.81, p = 0.37,Cohen’s f =.07), LCD (F(1,169) = 0.002, p =0.96, Cohen’s f = 0.05), SWM (F(1,171) = 1.05, p = 0.31,Cohen’s f = 0.08).

EEG Predictors of WM RT and SDRT

The results of regressing composite reaction timemeasures (RT/SDRT) on EEG features, across core WM functions, is shownin Table 5. Lower composite RT/SDRT scores (i.e., faster andless variable responding) were associated with higher P3b am-plitudes (p = 0.006), as well as more negative (and thereforestronger) CNV in the DPX task (p = 0.03). Thus, composite RT/RTSD scores were most strongly associated with EEG indica-tors of WM updating and goal maintenance, respectively.

Load Effects The P3b and CNV were also examined for loadeffects. The amplitude of P3b at electrode Pz during encoding

Fig. 3 Event-related changes in spectral power are shown for each task,across the trial window, expressed in dB scale relative to the prestimulusbaseline. The plots show responses for independent components withoccipital topography. Inset at center shows this topography, with redindicating electrodes maximally contributing to the IC timeseries. Theduration of the encoding target and probe stimulus is indicated with ablack bar in each plot. The top two rows show exemplary low-load andhigh-load conditions for non-care-seeking (NCS) individuals, the bottom

two rows show exemplary low-load and high-load conditions for care-seeking (CS) individuals. Clear decreases in power are apparent duringeach stimulus across 8-20 Hz (blue), with the DFR task also showingincreases in power during the maintenance period. These EEG featuresalso show clear load (greater at high load) and group (more negative inNCS) effects. DFR = delayed face recognition, SWM = Sternberg spatialworking memory; LCD = lateralized change detection; DPX = dot-pattern expectancy


showed significant load effects, although the direction variedwith task. Amplitude decreased with increasing load in theDFR task, F(1,196) = 9.5, p = 0.002, Cohen’s f = 0.22 (Mlow to high = 4.2, 3.8 uV). In contrast, amplitude increasedwith increasing load in the LCD task, F(1,193) = 5.3, p =0.005, Cohen’s f = 0.17 (M low to high = 3.8, 4.1, 4.2 uV) ,and the SWM task,F(1,197) = 4.4, p = 0.004,Cohen’s f = 0.15(M low to high = 2.0, 2.6, 2.4, 2.4 uV). This reversal is likelydue to known differences in task parameters (c.f., Discussion:WM Task Differences). In the DPX task, load was not manip-ulated directly; hence, instead, we evaluated effects of cuetype on P3b and, also, CNV amplitude, expecting that A cueswould increase demand for WM maintenance relative to B

cues. The P3b amplitude, consistent with the load effects inDFR, decreased in amplitude with increasing load (M B-cues= 5.4 uV,M A-cues = 3.2 uV), F(1,192) = 115.4, p = 0.0001,Cohen’s f = 0.78. Amplitude of CNV was as expected, largerfor A-cues than B-Cues (M B-cues = 0.08 uV, M A-cues =−0.58 uV), F(1,192) = 21.6, p = 0.0001, Cohen’s f = 0.34. Insum, load-effect analyses provide convergent evidence for therole of P3b, and CNV in WM RT/SDRT.

Group Effects Significant effects were observed in both CNVand P3b across all tasks, with the CS group showing consistentlyweaker CNV/P3b than the NCS group: CNV (F(1,192) = 6.7, p= 0.01, Cohen’s f = 0.19) (CS: M = −0.11 uV, SE = 0.06; NCS:

Fig. 4 Event-related changes in spectral power are shown for each task,across the trial window, expressed in dB scale relative to the prestimulusbaseline. The plots show responses for independent components withfrontal topography. Inset at center shows this topography, with redindicating electrodes maximally contributing to the IC timeseries. Theduration of the encoding target and probe stimulus is indicated with ablack bar in each plot. The top two rows show exemplary low-load and

high-load conditions for non-care-seeking (NCS) individuals, the bottomtwo rows show exemplary low-load and high-load conditions for care-seeking (CS) individuals. Stimulus events (black bars) are synchronouswith power increases (4-8 Hz) and decreases (15-20 Hz) that vary indegree across tasks and sampling groups. DFR = delayed face recogni-tion; SWM = Sternberg spatial working memory; LCD = lateralizedchange detection; DPX = dot-pattern expectancy.


M = −0.39 uV, SE = 0.09); P3b in DPX (F(1,192) = 3.6, p =0.06, Cohen’s f = 0.14) (CS: M = 4.0 uV, SE = 0.17; NCS: M =4.6 uV, SE = 0.26); P3b in DFR (F(1,196) = 3.2, p = 0.08,Cohen’s f = 0.13) (CS: M = 3.6 uV, SE = 0.23; NCS: M = 4.4uV, SE = .35); P3b in LCD (F(1,193) = 8.0, p=.005,Cohen’s f =0.20) (CS:M = 3.5 uV, SE = 0.19;NCS:M = 4.5 uV, SE = 0.30);and P3b in SWM (F(1,197) = 3.4, p = 0.065, Cohen’s f =.14)(CS: M = 2.1 uV, SE = 0.14; NCS: M = 2.6 uV, SE = 0.21).

Clinical Covariates

Next, we examined the robustness of the significant EEGpredictors of WM capacity and RT/RTSD to clinical

covariates. We re-ran the multiple regression analyses includ-ing one of: WHODAS overall disability index (Model A),depression and anxiety subscale theta scores for the BRIEFscales (Model B), or depression and anxiety scores for thePROMIS scales (Model C). The significant predictors in bothRT/RTSD and capacity analyses were unchanged across allthree models in all but one case.

Predictors of WM Capacity The EEG regression coefficientfor CDA remained significant across all three models(Model A: t(153) = −2.5, p = 0.005; Model B: t(153) =−2.4, p = 0.02; Model C: t(153) = −2.0, p = 0.05). TheEEG regression coefficient for frontal theta remained

Fig. 5 Event-related changes in potential are shown for each task, acrossthe trial window, expressed relative to the prestimulus baseline. Plotsshown for select electrodes. The top two rows show P3b responses fornon-care-seeking (NCS) and care-seeking (CS) individuals across all fourtasks. Bottom two rows showN170, CDA and CNVERPs in DFR, LCD,and DPX respective, for NCS and CS individuals. The P3b is greater forNCS participants and shows expected decreases with load in the DFRtask, but increased amplitude with load in SWM, LCD andDPX, possibly

due to confounded increases in saliency. The N170 in DFR is also en-hanced with higher load. The CDA is evident in the LCD task, with morenegative potential (e.g., 0.2-0.6 s poststimulus) for contralateral stimuli ineach hemisphere (r = right, l = left). The CNV in DPX task, at electrodeFCz, is more negative in AX than BX trials consistent with expectancy ofX stimuli and associated responses. DFR = delayed face recognition;SWM = Sternberg spatial working memory; LCD = lateralized changedetection; DPX = dot-pattern expectancy


significant across Models A and B models (Model A:t(153) = 2.5, p = 0.01; Model B: t(153) = 2.6, p = 0.01)but not Model C (t(153) = 1.5, p = 0.1).

Predictors of RT/RTSD The EEG regression coefficient forCNV remained significant across all three models(Model A: t(153) = 2.0, p = 0.05; Model B: t(153) =2.1, p = 0.04; Model C: t(153) = 2.1, p = 0.03). TheEEG regression coefficient for P3b also remained sig-nificant across all three models (Model A: t(153) =−2.4, p = 0.02; Model B: t(153) = −2.9, p = 0.005;Model C: t(153) = −2.5, p = 0.02).

Additionally, the regression coefficients for covariatepredictors were not significant, suggesting the observedrelationships between EEG features and WM capacity orRT/RTSD are not accounted for by shared variance be-tween EEG features and clinical covariates. The first-order correlations between EEG features and clinicalcovariates were largely not significant (Table S5), withthe exception of P3b and less reliably for CDA andCNV, all of which weakened in magnitude as indicatorsof anxiety and depression increased.

First-Order Correlations

Finally, to further support interpretation of the aboverelationships, we evaluated the first-order correlationsbetween composite outcome measures and the sevenEEG predictors. As shown in Table 6 and Figure 6,first-order correlations were significant for significantpredictors of composite capacity (CDA, theta) and com-posite RT/SDRT (P3b, CNV). Correlations among pre-dictors were scarce with the exception of occipital al-pha, which correlated negatively with P3b amplitude(r(187) = −0.24, p = 0.001), and with maintenance theta(r(153) = 0.17, p = 0.04), suggesting that this measuremay play an additional indirect role in WM processes.

Discussion

In this study we sought to test the relative contribution ofputative WM processes, as measured by EEG features acrossfour WM tasks, to WM performance. We found that storagesub-processes during maintenance (theta, CDA) were relatedto WM capacity, and were distinct from updating (P3b) andgoal maintenance (CNV) processes, which were more strong-ly associated with reaction time measures. The results validatethese putative subprocesses of WM and suggest that the mostsalient distinction in WM mechanisms lies in tonic processesduring maintenance, versus phasic processes associated withstimulus processing. In addition, the findings highlight varia-tions in the ability of WM tasks to measure different subpro-cesses and differences in reliability and validity of EEG mea-sures, ranging from high (e.g., P3b, alpha ERD) to low (CFC,gamma).

Subprocesses of WM

One contribution of the present work is in validating coresubprocesses of WM, the identity and separability of whichhave been the subject of debate. For instance, the NIH RDoCframework (Insel et al., 2010) has proposed four putative do-mains of WM that include active maintenance, limited capac-ity, flexible updating, and interference control. Our resultspartially validate these dimensions and additionally suggestthat active maintenance and limited capacity are coupledthrough their relationship to behavioral WM capacity mea-sures, whereas updating and goal maintenance processes to-gether are more associated with response speed and variabil-ity. The association of updating processes and response vari-ability is notable, due to the potential role of such processes incontributing to WM impairments in ADHD (Lenartowiczet al., 2018; Lenartowicz et al., 2019). A caveat to our results,however, is that the chosen tasks were not designed to isolateWM updating processes from concurrent operations, such as

Table 5 Multiple regression: EEG predictors of working memory performance

Construct Model predictors DV: Capacity composite scores DV: RT/RTSD composite scores

β tβ β tβ

Short-term storage CDA amplitude −0.21 −2.72** −0.07 0.86

N170 amplitude (O1 & O2) 0.07 0.86 0.02 0.26

Long-term storage PC1: Frontal Theta 0.22 2.79** 0.04 0.49

Goal maintenance CNV amplitude (FCz) -0.15 −1.91† 0.16 2.02*

WM updating PC1: composite posterior P3b 0.16 1.92† −0.22 −2.68**

PC1: composite posterior alpha ERD −0.05 0.60 0.10 1.23

Alerting P2 amplitude (Fz) 0.03 0.40 0.06 0.76

Model Fit R2adj = 0.17, F = 2.91***, n = 153 R2

adj = 0.10, F = 2.64**, n = 153

†p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001; ~ indicates that tβ < 1; DV = model dependent variable


stimulus categorization or automatic updating (Kessler, 2017;Rac-Lubashevsky & Kessler, 2016), and novel measures maybe needed to provide greater sensitivity to these distinctions.Given that, in the service of dimension reduction, we col-lapsed across encoding and retrieval operations, it is likelythat our results reflect general aspects of WM updating thatinvolve interactions between sensory and associative systems.

In addition, some have debated whether the distinction be-tween short-term storage (STM) and long-term storage (LTM)systems in WM is valid (Ranganath & Blumenfeld, 2005;Ruchkin, Grafman, Cameron, & Berndt, 2003), with severallines of research examining whether necessary conditions ex-ist for the recruitment of cortico-hippocampal LTM mecha-nisms during WM maintenance (Axmacher et al., 2007;Rissman et al., 2008; Sakai & Passingham, 2004). Our resultssuggest that, across four tasks of variable demands, both LTM(theta) and STM (CDA) mechanisms are recruited duringmaintenance and both contribute to capacity, which broadlysupports the view that minimizes distinctions between thesesystems in WM maintenance.

The results of this study may additionally support a simplertwo-dimensional view of WM, namely the interplay betweentonic processes that seek to maintain the stability of activatedstates free from interference from exogenous stimuli, and pha-sic processes that lead to shifting of activation states in re-sponse to novel stimuli and updating of response processing(Aston-Jones & Cohen, 2005; Bilder, 2012; Bilder, Volavka,Lachman, & Grace, 2004; Pribram & McGuinness, 1975).One of the identified behavioral dimensions coupled twomaintenance-related features (frontal theta and CDA) and per-formance capacity scores, consistent with tonic processes. Incontrast, the second dimension coupled P3b and CNV withreaction time metrics. The P3b is a putative correlate of WMupdating (Polich, 2007) and, in alternate formulations, of cat-egorization of content within WM (Kok, 2001; Nieuwenhuiset al., 2005). Thus, the P3b, such as the RT/RTSDmeasures, isconsistent with phasic processes, during which sensory inputstrigger neural interactions and phasic reorganization of

activity in the service of some outcome, like a response. Theassociation of CNV with this dimension may appear contra-dictory, given that CNV was introduced as a correlate of goalmaintenance. However, the CNV as we recorded it, is actuallyan ERP thought to index a combination of goal maintenanceprocesses, and also, response preparation processes (Babiloniet al., 2005; Hamano et al., 1997). Thus, like the P3b, it is anindicator of phasic neurophysiological responses. From thisperspective, the analysis casts WM as a function of tonicand phasic interactions that drive capacity and response out-comes, respectively.

Exploratory analyses showed that the behavioral assays ofWM and the EEG measures acquired during WM tasks wererelated to clinical state variables in predicted ways; indeed,both the cognitive task measures and factors, and the EEGmeasures all differed between care-seeking (CS) and non-care-seeking (NCS) groups, with effect sizes in the “medium”range. We also noted modest correlations of both behavioraland EEG indicators with individual clinical variables. The P3b“updating” EEG measures appeared to have a broader impactacross measures of clinical symptoms spanning depressionand anxiety, while the CDA measure had more circumscribedassociations with disability. While our study was not designedto address the hypothesis that there may be distinctive patternsof association between clinical symptoms and these WM in-dicators, future research might target these distinctions.

Finally, we highlight that a limitation of the study is a lackof non-WM control tasks. The absence of these control taskslimits interpretation of, in particular, the RT/RTSD factor,which may be a generic speed-related factor, not WM-related.Indeed, the dissociation of RT/RTSD from other accuracy-based indices has been reported previously in the context ofclinical assessments in ADHD (Frazier-Wood et al., 2012;Kofler et al., 2014; Michelini et al., 2018). The present resultsexpand on these observations by tying RT/RTSD with phasicneural responses, here measured by P3b and CNV. Similarly,the interpretation of our results is limited to the domain ofspatial working memory, as verbal stimuli were excluded to

Table 6 Variable correlations

1 2 3 4 5 6 7 8 9

1. Capacity composite scores Performance

2. RT/RTSD composite scores −0.41**3. CDA amplitude −0.15* −0.01 STM

4. N170 amplitude (O1 & O2) 0.01 0.01 0.01

5. PC1: Frontal Theta 0.22** 0.04 0.03 −0.06 LTM

6. CNV amplitude (FCz) −0.14 0.21** 0.06 0.12 −0.02 GM

7. PC1: composite posterior P3b 0.22** −0.29** 0 0.01 0 −0.07 Updating

8. PC1: composite posterior alpha ERD −0.06 0.08 0.02 −0.13 0.17* −0.07 −0.24**9. P2 amplitude (Fz) 0.08 −0.05 −0.04 −0.08 −0.12 −0.12 0.03 −0.13 Alerting

STM = short-term storage, LTM = long-term storage, GM = goal maintenance. †p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001


control for individual differences in semantic knowledge.Additionally, we note that while certain task conditions areparticularly important in assessing certain sub-processes(e.g., the DPX task was designed specifically to assess goalmaintenance), it should be acknowledged that all of the tasksrequire all processes to some extent in order to succeed.

EEG Metrics of WM

This study also highlights variability in reliability of EEGfeatures previously associated with WM subprocesses. TheP3b and alpha ERD both showed strong across-task reliabili-ty, comparable to that of WM performance across tasks, andreplicated known load effects and group effects. The latteralso was the case for CDA reported in the LCD task, andCNV reported in CPT paradigms. In contrast, gamma andtheta showed weaker reliability (Cronbach’s alpha ~ 0.5),whereas CFC measures were unreliable (Cronbach’s alpha= 0.13), and across all these measures load and group effectsfailed to replicate. The underperformance of the CFC is likelycoupled with weak reliability of gamma, reflected in weakacross-task Cronbach’s alpha, an absence of load effectsand, relatedly, non-parsimonious result in PCA on these met-rics. Assessment of gamma at the scalp is notoriously difficult(Fries, Scheeringa, & Oostenveld, 2008; Lopes da Silva,2013; Muthukumaraswamy, 2013) due to strong attenuationof high-frequency amplitudes by the scalp, as well as suscep-tibility of gamma-range frequencies to muscle generated arti-facts (e.g., micro-saccades) (Yuval-Greenberg & Deouell,2009; Yuval-Greenberg, Tomer, Keren, Nelken, & Deouell,2008). Indeed, some of the most robust reports of gramma-range CFC findings were reported from intracortical record-ings (Canolty et al., 2006; van Vugt, Schulze-Bonhage, Litt,Brandt, &Kahana, 2010; Voytek et al., 2010). Relatedly, priorreports of gamma modulation in scalp EEG recordings per-haps benefited from larger trial counts (e.g., n = 150, Roux &Uhlhaas, 2014; Roux, Wibral, Mohr, Singer, & Uhlhaas,2012), than employed in the current study (n = 20-40 percondition). We conclude therefore that gamma range oscilla-tions were not reliable in the current dataset due to limitationsof EEG inmeasuring high frequencies and low power, not thatthese measures are poor indices of local cortical processing.

In contrast to gamma, theta-range power, a proposed indexof long-term storage and/or cortico-hippocampal interactions,while also failing to show strong across-task reliability(Cronbach’s alpha = 0.5), did produce a parsimonious singlecomponent in principal component reduction and did showload effects, albeit inconsistently so. Theta power was alsopredictive of WM capacity, consistent with predictions. Theresults are therefore suggestive but lacking convergence. One

a

b

�Fig. 6 Scatterplots for first order correlations between compositebehavioral outcome scores and EEG predictor variables that showedsignificant regression coefficient effects in multiple regression analysis(Table 5). a) Correlations with capacity composite principal component(PC) scores. Greater capacity was associated with more negative CDA inLCD task, and also greater theta power during maintenance, across alltasks. b) Correlations with RT/RTSD composite PC scores. Faster andless variable responses were associated with larger P3b amplitude duringstimulus processing, across tasks, and more negative CNV in the 100 mspreceding X-stimuli in the DPX task. *p < 0.05; **p < 0.01


possibility is that the chosen WM tasks did not generate theputative conditions (e.g., novelty, relational encoding, disrup-tion of rehearsal, supra-capacity loads) posited as necessaryfor the recruitment of cortico-hippocampal LTM mechanismsduring WM maintenance (Axmacher et al., 2007; Rissmanet al., 2008; Sakai & Passingham, 2004). In some support ofthis hypothesis, the only task to show elevated power in thelow alpha, high theta band during maintenance was the DFRtask (Figure 3), which also was one of the most difficult (withrespect to accuracy) and the only task that required mainte-nance of multiple complex objects (e.g., faces vs. spatial fea-tures). One surprising finding was that alpha ERD, despitestrong reliability across tasks, did not correlate with WM per-formance, in contrast to prior reports (Lenartowicz et al.,2018). However, a key difference in the present sample wasthe sparsity of attentional deficits, suggesting that alpha ERD,and its associated attentional processes during stimulusencoding or retrieval, make only indirect contributions toWM performance that may manifest in performance whendeficits arise. It is notable that alpha ERD was the only mea-sure in the current analysis that was also correlated with otherpredictors of WM performance (e.g., P3b, Theta; Table 6),consistent with this prediction.

WM Task Differences

A final contribution of the present study, the results dem-onstrate how WM can tasks differ in relative recruitment ofdifferent WM subprocesses. While all four tasks showedgroup differences (CS vs. NCS) in RT/RTSD, differencesin measures of capacity were more pronounced in DPX andLCD tasks, than in DFR and SWM tasks. These observa-tions imply that, first, phasic, stimulus-related processesassociated with RT/RTSD, are reliable across tasks, alsoconsistent with high across-task reliability measures forP3b and alpha ERD. Second, demands on capacity maydiffer between DPX/LCD and DFR/SWM tasks. One dif-ference between these pairs is the task pace (Figure 1), withthe former be ing re la t ive ly fas t paced (<2-secmaintenance) and the latter being relatively slower paced(3-sec maintenance). Another difference is in type ofencoding, with the latter being delay-match to sampletasks, and the former including change-detection (LCD)and matching to an internal template (DPX). Either ofthese differences could imply a greater sensitivity ofDPX/LCD tasks to limited capacity or, in complement, agreater demand of SWM/DFR tasks on recruitment ofcomplementary storage mechanisms, like cortico-hippocampal circuitry. As noted above, the DFR task isunique in being the only task to show elevated alpha/theta power during maintenance, along with lower accura-cy, and increased complexity of to-be-stored objects.Consistent with this proposal, it is also the only task to

show decreased P3b amplitude with load, previously sug-gested to arise with a change in encoding mechanisms withelevated task difficulty. In contrast, in SWM and LCD, P3bincreased as load increased greater salience, a phenomenonconsistent with prior observations of the P3b potential(Kok, 2001). Thus, we would hypothesize that DFR re-cruits LTM processes to a greater extent than spatialencoding (DPX, SWM) or dot-pattern recognition (DPX).

Conclusions

The objective of this study was to test the relative contributionof putativeWMprocesses, as measured by EEG features acrossfour WM tasks, to WM performance. The results validate thecontribution of short- and long-term maintenance processes(measured by CDA and theta) toWM capacity, in juxtapositionto the contribution of goal maintenance (CNV) and updating(P3b) to RT/RTSD. We suggest that the findings are consistentwith a dual-process model of WM that sees variation in WMfunctions as the outcomes of balance between underlying tonicand phasic neurophysiological processes. Future studies will beneeded to validate the strength of the relationship between theidentified EEG metrics and underlying, putative circuitry ofeach WM process. The results additionally highlight thatLCD and DPX tasks were more sensitive to capacity groupdifferences, and therefore perhaps better suited to assessingrelated neural processes, and the DFR task perhaps better suitedto long-term storage assessment. Finally, there exist clear dif-ferences in the across-task internal consistency of EEG features,with strongest consistency for P3b and alpha ERD (on par withreliability of performance across tasks), and negligible consis-tency for CFC metrics.

Supplementary Information The online version contains supplementarymaterial available at https://doi.org/10.3758/s13415-021-00924-7.

Acknowledgments This work was supported by funding from theNational Institutes of Mental Health (R01MH101478; R01MH118514)to R.M.B., and to A.L. and S.K.L (R01MH116268), as well as fromBrain& Behavior Research Foundation, Young Investigator grant to A.L.

Open Access This article is licensed under a Creative CommonsAttribution 4.0 International License, which permits use, sharing, adap-tation, distribution and reproduction in any medium or format, as long asyou give appropriate credit to the original author(s) and the source, pro-vide a link to the Creative Commons licence, and indicate if changes weremade. The images or other third party material in this article are includedin the article's Creative Commons licence, unless indicated otherwise in acredit line to the material. If material is not included in the article'sCreative Commons licence and your intended use is not permitted bystatutory regulation or exceeds the permitted use, you will need to obtainpermission directly from the copyright holder. To view a copy of thislicence, visit http://creativecommons.org/licenses/by/4.0/.


https://doi.org/10.3758/s13415-021-00924-7

https://doi.org/

References

Alexander, M. P., Stuss, D. T., Picton, T., Shallice, T., & Gillingham, S.(2007). Regional frontal injuries cause distinct impairments in cog-nitive control. Neurology, 68(18), 1515-1523. Retrieved from <Goto ISI>://000246130800013

Aston-Jones, G., & Cohen, J. D. (2005). An integrative theory of locuscoeruleus-norepinephrine function: Adaptive gain and optimal per-formance. Annual Review of Neuroscience, 28, 403-450. https://doi.org/10.1146/Annurev.Neuro.28.061604.135709

Axmacher, N., Henseler, M.M., Jensen, O.,Weinreich, I., Elger, C. E., &Fell, J. (2010). Cross-frequency coupling supports multi-item work-ing memory in the human hippocampus. Proceedings of theNational Academy of Sciences of the United States of America,107(7), 3228-3233. https://doi.org/10.1073/pnas.0911531107

Axmacher, N., Mormann, F., Fernandez, G., Cohen, M. X., Elger, C. E.,& Fell, J. (2007). Sustained neural activity patterns during workingmemory in the human medial temporal lobe. The Journal ofNeuroscience, 27(29), 7807-7816. https://doi.org/10.1523/JNEUROSCI.0962-07.2007

Babiloni, C., Brancucci, A., Capotosto, P., Romani, G. L., Arendt-Nielsen, L., Chen, A. C., & Rossini, P. M. (2005). Slow corticalpotential shifts preceding sensorimotor interactions. Brain ResearchBulletin, 65(4), 309-316. https://doi.org/10.1016/j.brainresbull.2004.11.023

Baddeley, A. (1986). Working Memory. Oxford University Press.Baddeley, A. (2002). Is working memory still working? European

Psychologist, 7(2), 85-97. Retrieved from http://www.apa.org/journals/epp.html, http://www.hogrefe.de

Barch, D. M., Berman, M. G., Engle, R., Jones, J. H., Jonides, J.,Macdonald, A., 3rd, . . . Sponheim, S. R. (2009a). CNTRICS finaltask selection: working memory. Schizophrenia Bulletin, 35(1),136-152. https://doi.org/10.1093/schbul/sbn153

Barch, D. M., Carter, C. S., Arnsten, A., Buchanan, R. W., Cohen, J. D.,Geyer, M., . . . Heinssen, R. (2009b). Selecting paradigms fromcognitive neuroscience for translation into use in clinical trials: pro-ceedings of the third CNTRICS meeting. Schizophrenia Bulletin,35(1), 109-114. https://doi.org/10.1093/schbul/sbn163

Barch, D. M., & Dowd, E. C. (2010). Goal representations and motiva-tional drive in schizophrenia: the role of prefrontal-striatal interac-tions. Schizophrenia Bulletin, 36(5), 919-934. https://doi.org/10.1093/schbul/sbq068

Barch, D. M., Mitropoulou, V., Harvey, P. D., New, A. S., Silverman, J.M., & Siever, L. J. (2004). Context-processing deficits inschizotypal personality disorder. Journal of Abnormal Psychology,113(4), 556-568. Retrieved from http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retr ieve&db=PubMed&dopt=Citation&list_uids=15535788

Barch, D. M., Moore, H., Nee, D. E., Manoach, D. S., & Luck, S. J.(2012). CNTRICS imaging biomarkers selection: Working memo-ry. Schizophrenia Bulletin, 38(1), 43-52. https://doi.org/10.1093/schbul/sbr160

Bearden, C. E., Hoffman, K. M., & Cannon, T. D. (2001). The neuropsy-chology and neuroanatomy of bipolar affective disorder: a criticalreview. Bipolar Disorders, 3(3), 106-150; discussion 151-103.https://doi.org/10.1034/j.1399-5618.2001.030302.x

Bechara, A., &Martin, E. M. (2004). Impaired decisionmaking related toworking memory deficits in individuals with substance addictions.Neuropsychology, 18(1), 152-162. https://doi.org/10.1037/0894-4105.18.1.152

Bilder, R.M. (2012). Executive control: balancing stability and flexibilityvia the duality of evolutionary neuroanatomical trends.Dialogues inClinical Neuroscience, 14(1), 39-47. Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/22577303

Bilder, R. M., Lenartowicz, A., Rissman, J., Loo, S., Pochon, J. B.,Enriquez, K., . . .… Hellemann, G. (2019). Spanning levels in theRDoC Matrix: Does working memory work? Paper presented at theSociety for Biological Psychiatry, Chicago.

Bilder, R. M., Volavka, J., Lachman, H. M., & Grace, A. A. (2004). Thecatechol-O-methyltransferase polymorphism: relations to the tonic-phasic dopamine hypothesis and neuropsychiatric phenotypes.Neuropsychopharmacology, 29(11), 1943-1961. https://doi.org/10.1038/sj.npp.1300542

Brewin, C. R., Gregory, J. D., Lipton,M., &Burgess, N. (2010). Intrusiveimages in psychological disorders: characteristics, neural mecha-nisms, and treatment implications. Psychological Review, 117(1),210-232. https://doi.org/10.1037/a0018113

Canolty, R. T., Edwards, E., Dalal, S. S., Soltani, M., Nagarajan, S. S.,Kirsch, H. E.,…Knight, R. T. (2006). High gamma power is phase-locked to theta oscillations in human neocortex. Science, 313(5793),1626-1628. https://doi.org/10.1126/science.1128115

Cohen, J. (1988). Statistical power analysis for the behavioral sciences. :Routledge Academic.

Cohen, J. D., Barch, D. M., Carter, C., & Servan-Schreiber, D. (1999).Context-processing deficits in schizophrenia: converging evidencefrom three theoretically motivated cognitive tasks. Journal ofAbnormal Psychology, 108(1), 120-133. Retrieved from http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=10066998

Compte, A., Brunel, N., Goldman-Rakic, P. S., & Wang, X. J. (2000).Synaptic mechanisms and network dynamics underlying spatialworking memory in a cortical network model. Cerebral Cortex,10(9), 910-923. https://doi.org/10.1093/cercor/10.9.910

Cowan, N. (2008). What are the differences between long-term, short-term, and working memory? Progress in Brain Research, 169, 323-338. https://doi.org/10.1016/S0079-6123(07)00020-9

D'Ardenne, K., Eshel, N., Luka, J., Lenartowicz, A., Nystrom, L. E., &Cohen, J. D. (2012). Role of prefrontal cortex and the midbraindopamine system in working memory updating. Proceedings ofthe National Academy of Sciences of the United States of America,109(49), 19900-19909. https://doi.org/10.1073/Pnas.1116727109

D'Esposito, M., Aguirre, G. K., Zarahn, E., Ballard, D., Shin, R. K., &Lease, J. (1998). Functional MRI studies of spatial and nonspatialworking memory. Cognitive Brain Research, 7, 1-13.

Daume, J., Gruber, T., Engel, A. K., & Friese, U. (2017). Phase-Amplitude Coupling and Long-Range Phase SynchronizationReveal Frontotemporal Interactions during Visual WorkingMemory. The Journal of Neuroscience, 37(2), 313-322. https://doi.org/10.1523/JNEUROSCI.2130-16.2016

Delorme, A., &Makeig, S. (2004). EEGLAB: an open source toolbox foranalysis of single-trial EEG dynamics including independent com-ponent analysis. Journal of Neuroscience Methods, 134(1), 9-21.Doi https://doi.org/10.1016/J.Jneumeth.2003.10.009

Foxe, J. J., & Snyder, A. C. (2011). The Role of Alpha-Band BrainOscillations as a Sensory Suppression Mechanism during SelectiveAttention. Frontiers in Psychology, 2, 154. https://doi.org/10.3389/fpsyg.2011.00154

Frazier-Wood, A. C., Bralten, J., Arias-Vasquez, A., Luman, M.,Ooterlaan, J., Sergeant, J., . . . Rommelse, N. N. (2012).Neuropsychological intra-individual variability explains unique ge-netic variance of ADHD and shows suggestive linkage to chromo-somes 12, 13, and 17. American Journal of Medical Genetics. PartB, Neuropsychiatric Genetics, 159B(2), 131-140. https://doi.org/10.1002/ajmg.b.32018

Fries, P., Scheeringa, R., & Oostenveld, R. (2008). Finding gamma.Neuron, 58(3), 303-305. https://doi.org/10.1016/j.neuron.2008.04.020

Fuster, J. M. (1985). The prefrontal cortex, mediator of cross-temporalcontingencies.HumanNeurobiology, 4(3), 169-179. Retrieved fromhttps://www.ncbi.nlm.nih.gov/pubmed/3934116


https://doi.org/10.1146/Annurev.Neuro.28.061604.135709

https://doi.org/10.1146/Annurev.Neuro.28.061604.135709

https://doi.org/10.1073/pnas.0911531107

https://doi.org/10.1523/JNEUROSCI.0962-07.2007


https://doi.org/10.1016/j.brainresbull.2004.11.023

https://doi.org/10.1016/j.brainresbull.2004.11.023

http://www.apa.org/journals/epp.html

http://www.apa.org/journals/epp.html

https://doi.org/10.1093/schbul/sbn153


https://doi.org/10.1093/schbul/sbq068

https://doi.org/10.1093/schbul/sbq068

http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=15535788



https://doi.org/10.1093/schbul/sbr160


https://doi.org/10.1034/j.1399-5618.2001.030302.x

https://doi.org/10.1037/0894-4105.18.1.152

https://doi.org/10.1037/0894-4105.18.1.152

https://www.ncbi.nlm.nih.gov/pubmed/22577303


https://doi.org/10.1038/sj.npp.1300542

https://doi.org/10.1038/sj.npp.1300542

https://doi.org/10.1037/a0018113

https://doi.org/10.1126/science.1128115




https://doi.org/10.1093/cercor/10.9.910

https://doi.org/10.1016/S0079-6123(07)00020-9

https://doi.org/10.1073/Pnas.1116727109



https://doi.org/10.1016/J.Jneumeth.2003.10.009

https://doi.org/10.3389/fpsyg.2011.00154


https://doi.org/10.1002/ajmg.b.32018

https://doi.org/10.1002/ajmg.b.32018

https://doi.org/10.1016/j.neuron.2008.04.020



Fuster, J. M., & Alexander, G. E. (1971). Neuron activity related to short-term memory. Science, 173(997), 652-654. Retrieved from http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=4998337

Glahn, D. C., Kim, J., Cohen, M. S., Poutanen, V. P., Therman, S., Bava,S., … Cannon, T. D. (2002). Maintenance and manipulation inspatial working memory: Dissociations in the prefrontal cortex.NeuroImage, 17(1), 201-213. Doi https://doi.org/10.1006/Nimg.2002.1161

Gold, J. M., Hahn, B., Zhang, W. W., Robinson, B. M., Kappenman, E.S., Beck, V. M., & Luck, S. J. (2010). Reduced capacity but sparedprecision and maintenance of working memory representations inschizophrenia. Archives of General Psychiatry, 67(6), 570-577.https://doi.org/10.1001/archgenpsychiatry.2010.65

Goldman-Rakic, P. S. (1995). Cellular basis of working memory.Neuron, 14(3), 477-485. https://doi.org/10.1016/0896-6273(95)90304-6

Griffiths, B. J., Parish, G., Roux, F., Michelmann, S., van der Plas, M.,Kolibius, L. D., .… Hanslmayr, S. (2019). Directional coupling ofslow and fast hippocampal gamma with neocortical alpha/beta os-cillations in human episodic memory. Proceedings of the NationalAcademy of Sciences of the United States of America, 116(43),21834-21842. https://doi.org/10.1073/pnas.1914180116

Haatveit, B. C., Sundet K Fau-Hugdahl, K., Hugdahl K Fau-Ueland, T.,Ueland T Fau-Melle, I., Melle I Fau-Andreassen, O. A., &Andreassen, O. A. (2010). The validity of d prime as a workingmemory index: results from the "Bergen n-back task". Journal ofClinical and Experimental Neuropsychology, 32(8), 871-880.

Hamano, T., Luders, H. O., Ikeda, A., Collura, T. F., Comair, Y. G., &Shibasaki, H. (1997). The cortical generators of the contingent neg-ative variation in humans: A study with subdural electrodes. EvokedPotentials-Electroencephalography and Clinical Neurophysiology,104(3), 257-268. Retrieved from <Go to ISI>://A1997XD24600008

Hanslmayr, S., Klimesch, W., Sauseng, P., Gruber, W., Doppelmayr, M.,Freunberger, R., & Pecherstorfer, T. (2005). Visual discriminationperformance is related to decreased alpha amplitude but increasedphase locking. Neuroscience Letters, 375(1), 64-68. https://doi.org/10.1016/j.neulet.2004.10.092

Henderson, D., Poppe, A. B., Barch, D. M., Carter, C. S., Gold, J. M.,Ragland, J. D., . . . MacDonald, A. W., 3rd. (2012). Optimization ofa goal maintenance task for use in clinical applications.Schizophrenia Bulletin, 38(1), 104-113. https://doi.org/10.1093/schbul/sbr172

Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D. S., Quinn, K., .. . Wang, P. (2010). Research domain criteria (RDoC): toward a newclassification framework for research on mental disorders. TheAmerican Journal of Psychiatry, 167(7), 748-751. https://doi.org/10.1176/appi.ajp.2010.09091379

Jensen, O., & Mazaheri, A. (2010). Shaping functional architecture byoscillatory alpha activity: gating by inhibition. Frontiers in HumanNeuroscience, 4, 186. https://doi.org/10.3389/fnhum.2010.00186

Jensen, O., & Tesche, C. D. (2002). Frontal theta activity in humansincreases with memory load in a working memory task. EuropeanJournal of Neuroscience, 15(8), 1395-1399. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/11994134

Joormann, J., & Gotlib, I. H. (2008). Updating the contents of workingmemory in depression: interference from irrelevant negative mate-rial. Journal of Abnormal Psychology, 117(1), 182-192. https://doi.org/10.1037/0021-843X.117.1.182

Kessler, Y. (2017). The Role of Working Memory Gating in TaskSwitching: A Procedural Version of the Reference-Back Paradigm.Frontiers in Psychology, 8, 2260. https://doi.org/10.3389/fpsyg.2017.02260

Kessler, Y., & Oberauer, K. (2015). Forward scanning in verbal workingmemory updating. Psychonomic Bulletin & Review, 22(6), 1770-1776. https://doi.org/10.3758/s13423-015-0853-0

Klimesch, W. (1997). EEG-alpha rhythms and memory processes.International Journal of Psychophysiology, 26(1-3), 319-340.Retrieved from <Go to ISI>://A1997XF06600022

Klimesch, W. (2012). Alpha-band oscillations, attention, and controlledaccess to stored information. Trends in Cognitive Sciences, 16(12),606-617. https://doi.org/10.1016/j.tics.2012.10.007

Kofler, M. J., Alderson, R. M., Raiker, J. S., Bolden, J., Sarver, D. E., &Rapport, M. D. (2014). Working memory and intraindividual vari-ability as neurocognitive indicators in ADHD: examining compet-ing model predictions. Neuropsychology, 28(3), 459-471. https://doi.org/10.1037/neu0000050

Kok, A. (2001). On the utility of P3 amplitude as a measure of processingcapacity. Psychophysiology, 38(3), 557-577. Retrieved from <Go toISI>://000168584400022

Lee, T. W., Girolami, M., & Sejnowski, T. J. (1999). Independent com-ponent analysis using an extended infomax algorithm for mixedsubgaussian and supergaussian sources. Neural Computation,11(2), 417-441. Retrieved from <Go to ISI>://000078883700005

Lenartowicz, A., Delorme, A., Walshaw, P. D., Cho, A. L., Bilder, R. M.,McGough, J. J., . . . Loo, S. K. (2014). Electroencephalographycorrelates of spatial working memory deficits in attention-deficit/hyperactivity disorder: vigilance, encoding, and maintenance. TheJournal of Neuroscience, 34(4), 1171-1182. https://doi.org/10.1523/JNEUROSCI.1765-13.2014

Lenartowicz, A., Escobedo-Quiroz, R., & Cohen, J. D. (2010). Updatingof context in working memory: An event-related potential study.Cognitive, Affective, & Behavioral Neuroscience, 10(2), 298-315.Doi https://doi.org/10.3758/Cabn.10.2.298

Lenartowicz, A., Mazaheri, A., Jensen, O., & Loo, S. K. (2018). AberrantModulation of Brain Oscillatory Activity and AttentionalImpairment in Attention-Deficit/Hyperactivity Disorder. BiolPsychiatry Cogn Neurosci Neuroimaging, 3(1), 19-29. https://doi.org/10.1016/j.bpsc.2017.09.009

Lenartowicz, A., Truong, H., Salgari, G. C., Bilder, R. M., McGough, J.,McCracken, J. T., & Loo, S. K. (2019). Alpha modulation duringworking memory encoding predicts neurocognitive impairment inADHD. Journal of Child Psychology and Psychiatry, 60(8), 917-926. https://doi.org/10.1111/jcpp.13042

Leonard, C. J., Kaiser, S. T., Robinson, B. M., Kappenman, E. S., Hahn,B., Gold, J. M., & Luck, S. J. (2013). Toward the neural mecha-nisms of reduced working memory capacity in schizophrenia.Cerebral Cortex, 23(7), 1582-1592. https://doi.org/10.1093/cercor/bhs148

Leszczynski, M., Fell, J., & Axmacher, N. (2015). Rhythmic WorkingMemory Activation in the Human Hippocampus. Cell Reports,13(6), 1272-1282. https://doi.org/10.1016/j.celrep.2015.09.081

Lisman, J. E., & Buzsaki, G. (2008). A neural coding scheme formed bythe combined function of gamma and theta oscillations.Schizophrenia Bulletin, 34(5), 974-980. https://doi.org/10.1093/schbul/sbn060

Lisman, J. E., & Idiart, M. A. (1995). Storage of 7 +/- 2 short-termmemories in oscillatory subcycles. Science, 267(5203), 1512-1515. https://doi.org/10.1126/science.7878473

Loo, S. K., Humphrey, L. A., Tapio, T., Moilanen, I. K., McGough, J. J.,McCracken, J. T., … Smalley, S. L. (2007). Executive functioningamong Finnish adolescents with attention-deficit/hyperactivity dis-order. Journal of the American Academy of Child and AdolescentPsychiatry, 46(12), 1594-1604. https://doi.org/10.1097/chi.0b013e3181575014

Lopes da Silva, F. (2013). EEG and MEG: relevance to neuroscience.Neuron, 80(5), 1112-1128. https://doi.org/10.1016/j.neuron.2013.10.017

Luck, S. J., & Vogel, E. K. (1997). The capacity of visual workingmemory for features and conjunctions. Nature, 390(6657), 279-281. https://doi.org/10.1038/36846





https://doi.org/10.1006/Nimg.2002.1161

https://doi.org/10.1006/Nimg.2002.1161

https://doi.org/10.1001/archgenpsychiatry.2010.65

https://doi.org/10.1016/0896-6273(95)90304-6

https://doi.org/10.1016/0896-6273(95)90304-6

https://doi.org/10.1073/pnas.1914180116

https://doi.org/10.1016/j.neulet.2004.10.092

https://doi.org/10.1016/j.neulet.2004.10.092



https://doi.org/10.1176/appi.ajp.2010.09091379

https://doi.org/10.1176/appi.ajp.2010.09091379

https://doi.org/10.3389/fnhum.2010.00186

http://www.ncbi.nlm.nih.gov/pubmed/11994134

http://www.ncbi.nlm.nih.gov/pubmed/11994134

https://doi.org/10.1037/0021-843X.117.1.182

https://doi.org/10.1037/0021-843X.117.1.182



https://doi.org/10.3758/s13423-015-0853-0

https://doi.org/10.1016/j.tics.2012.10.007

https://doi.org/10.1037/neu0000050

https://doi.org/10.1037/neu0000050



https://doi.org/10.3758/Cabn.10.2.298

https://doi.org/10.1016/j.bpsc.2017.09.009

https://doi.org/10.1016/j.bpsc.2017.09.009

https://doi.org/10.1111/jcpp.13042

https://doi.org/10.1093/cercor/bhs148

https://doi.org/10.1093/cercor/bhs148

https://doi.org/10.1016/j.celrep.2015.09.081



https://doi.org/10.1126/science.7878473

https://doi.org/10.1097/chi.0b013e3181575014

https://doi.org/10.1097/chi.0b013e3181575014



https://doi.org/10.1038/36846

Michelini, G., Cheung, C. H. M., Kitsune, V., Brandeis, D.,Banaschewski, T., McLoughlin, G., . . . Kuntsi, J. (2018). TheEtiological Structure of Cognitive-NeurophysiologicalImpairments in ADHD in Adolescence and Young Adulthood.Journal of Attention Disorders, 1087054718771191. https://doi.org/10.1177/1087054718771191

Miller, G. A. (1956). The magical number seven plus or minus two: somelimits on our capacity for processing information. PsychologyRevison, 63(2), 81-97. Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/13310704

Mitchell, D. J., McNaughton, N., Flanagan, D., & Kirk, I. J. (2008).Frontal-midline theta from the perspective of hippocampal "theta".Progress in Neurobiology, 86(3), 156-185. https://doi.org/10.1016/j.pneurobio.2008.09.005

Muthukumaraswamy, S. D. (2013). High-frequency brain activity andmuscle artifacts in MEG/EEG: a review and recommendations.Frontiers in Human Neuroscience, 7, 138. https://doi.org/10.3389/fnhum.2013.00138

Nee, D. E., & Jonides, J. (2011). Dissociable contributions of prefrontalcortex and the hippocampus to short-termmemory: evidence for a 3-state model of memory. NeuroImage, 54(2), 1540-1548. https://doi.org/10.1016/j.neuroimage.2010.09.002

Nee, D. E., Wager, T. D., & Jonides, J. (2007). Interference resolution:insights from a meta-analysis of neuroimaging tasks. Cognitive,Affective, & Behavioral Neuroscience, 7(1), 1-17. https://doi.org/10.3758/cabn.7.1.1

Nieuwenhuis, S., Aston-Jones, G., & Cohen, J. D. (2005). Decision mak-ing, the P3, and the locus coeruleus-norepinephrine system.Psychological Bulletin, 131(4), 510-532. Retrieved from http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=16060800

NIMH. (2011). Research Domains Criteria Initiative - WorkingMemory:Workshop Proceedings. Bethesda, Maryland, July 11-13 2010.Retrieved from https://www.nimh.nih.gov/research/research-funded-by-nimh/rdoc/working-memory-workshop-proceedings.shtml

O'Reilly, R. C., & Frank, M. J. (2006). Making working memory work: acomputational model of learning in the prefrontal cortex and basalganglia. Neural Computation, 18(2), 283-328. https://doi.org/10.1162/089976606775093909

Oberauer, K., Süß, H.-M., Schulze, R., Wilhelm, O., &Wittmann,W.W.(2000). Working memory capacity - facets of a cognitive abilityconstruct. Personality and Individual Differences, 29, 1017-1045.

Onton, J., Delorme, A., & Makeig, S. (2005). Frontal midline EEG dy-namics during working memory. NeuroImage, 27(2), 341-356. Doihttps://doi.org/10.1016/J.Neuroimage.2005.04.014

Polania, R., Paulus, W., & Nitsche, M. A. (2012). Noninvasivelydecoding the contents of visual working memory in the human pre-frontal cortex within high-gamma oscillatory patterns. Journal ofCognitive Neuroscience, 24(2), 304-314. https://doi.org/10.1162/jocn_a_00151

Polich, J. (2007). Updating p300: An integrative theory of P3a and P3bClinical Neurophysiology, 118(10), 2128-2148. Retrieved from<Go to ISI>://000250193200002

Poppe, A. B., Barch, D. M., Carter, C. S., Gold, J. M., Ragland, J. D.,Silverstein, S. M., & MacDonald, A. W., 3rd. (2016). ReducedFrontoparietal Activity in Schizophrenia Is Linked to a SpecificDeficit in Goal Maintenance: A Multisite Functional ImagingStudy. Schizophrenia Bulletin, 42(5), 1149-1157. https://doi.org/10.1093/schbul/sbw036

Potts, G. F. (2004). An ERP index of task relevance evaluation of visualstimuli. Brain and Cognition, 56(1), 5-13. Retrieved from <Go toISI>://000224313600002

Pribram, K. H., &McGuinness, D. (1975). Arousal, activation, and effortin the control of attention. Psychological Review, 82(2), 116-149.https://doi.org/10.1037/h0076780

Rac-Lubashevsky, R., & Kessler, Y. (2016). Dissociating working mem-ory updating and automatic updating: The reference-back paradigm.Journal of Experimental Psychology. Learning, Memory, andCognition, 42(6), 951-969. https://doi.org/10.1037/xlm0000219

Ranganath, C., & Blumenfeld, R. S. (2005). Doubts about double disso-ciations between short- and long-term memory. Trends in CognitiveSciences, 9(8), 374-380. https://doi.org/10.1016/j.tics.2005.06.009

Rissman, J., Gazzaley, A., & D'Esposito, M. (2008). Dynamic adjust-ments in prefrontal, hippocampal, and inferior temporal interactionswith increasing visual working memory load. Cerebral Cortex,18(7), 1618-1629. https://doi.org/10.1093/cercor/bhm195

Roberts, B. M., Hsieh, L. T., & Ranganath, C. (2013). Oscillatory activityduring maintenance of spatial and temporal information in workingmemory. Neuropsychologia, 51(2), 349-357. https://doi.org/10.1016/j.neuropsychologia.2012.10.009

Romei, V., Brodbeck, V., Michel, C., Amedi, A., Pascual-Leone, A., &Thut, G. (2008). Spontaneous fluctuations in posterior alpha-bandEEG activity reflect variability in excitability of human visual areas.Cerebral Cortex, 18(9), 2010-2018. Doi https://doi.org/10.1093/Cercor/Bhm229

Rosahl, S. K., & Knight, R. T. (1995). Role of prefrontal cortex in gen-eration of the contingent negative variation. Cerebral Cortex, 5(2),123-134. https://doi.org/10.1093/cercor/5.2.123

Rossion, B., & Corentin, J. (2011). The N170: understanding the time-course oof face perception in the human brain. In S. J. Luck (Ed.),The Oxford Handbook of Event-Related Potential Components (pp.115-142). Oxford University Press.

Roux, F., & Uhlhaas, P. J. (2014). Working memory and neural oscilla-tions: alpha-gamma versus theta-gamma codes for distinct WM in-formation? Trends in Cognitive Sciences, 18(1), 16-25. https://doi.org/10.1016/j.tics.2013.10.010

Roux, F., Wibral, M., Mohr, H. M., Singer, W., & Uhlhaas, P. J. (2012).Gamma-band activity in human prefrontal cortex codes for the num-ber of relevant itemsmaintained in workingmemory. The Journal ofNeuroscience, 32(36), 12411-12420. https://doi.org/10.1523/JNEUROSCI.0421-12.2012

Ruchkin, D. S., Grafman, J., Cameron, K., & Berndt, R. S. (2003).Working memory retention systems: a state of activated long-termmemory. The Behavioral and Brain Sciences, 26(6), 709-728; dis-cussion 728-777. https://doi.org/10.1017/s0140525x03000165

Sakai, K., & Passingham, R. E. (2004). Prefrontal selection and medialtemporal lobe reactivation in retrieval of short-term verbal informa-tion. Cerebral Cortex, 14(8), 914-921. https://doi.org/10.1093/cercor/bhh050

Samaha, J., & Postle, B. R. (2015). The Speed of Alpha-BandOscillations Predicts the Temporal Resolution of VisualPerception. Current Biology, 25(22), 2985-2990. https://doi.org/10.1016/j.cub.2015.10.007

Sauseng, P., Klimesch, W., Heise, K. F., Gruber, W. R., Holz, E., Karim,A. A., … Hummel, F. C. (2009). Brain oscillatory substrates ofvisual short-term memory capacity. Current Biology, 19(21),1846-1852. https://doi.org/10.1016/j.cub.2009.08.062

Stuss, D. T., & Benson, D. F. (1984). Neuropsychological studies of thefrontal lobes. Psychological Bulletin, 95(1), 3-28. Retrieved fromht tp: / /www.ncbi .nlm.nih.gov/entrez /query.fcgi?cmd=Retrieve&db=PubMed&dopt=Citation&list_uids=6544432

Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics,5th ed. Allyn & Bacon/Pearson Education.

Tanner, D., Morgan-Short, K., & Luck, S. J. (2015). How inappropriatehigh-pass filters can produce artifactual effects and incorrect conclu-sions in ERP studies of language and cognition. Psychophysiology,52(8), 997-1009. https://doi.org/10.1111/psyp.12437

Tanner, D., Norton, J. J., Morgan-Short, K., & Luck, S. J. (2016). Onhigh-pass filter artifacts (they're real) and baseline correction (it's agood idea) in ERP/ERMF analysis. Journal of Neuroscience


https://doi.org/10.1177/1087054718771191

https://doi.org/10.1177/1087054718771191



https://doi.org/10.1016/j.pneurobio.2008.09.005

https://doi.org/10.1016/j.pneurobio.2008.09.005



https://doi.org/10.1016/j.neuroimage.2010.09.002

https://doi.org/10.1016/j.neuroimage.2010.09.002

https://doi.org/10.3758/cabn.7.1.1

https://doi.org/10.3758/cabn.7.1.1




https://www.nimh.nih.gov/research/research-funded-by-nimh/rdoc/working-memory-workshop-proceedings.shtml



https://doi.org/10.1162/089976606775093909

https://doi.org/10.1162/089976606775093909

https://doi.org/10.1016/J.Neuroimage.2005.04.014

https://doi.org/10.1162/jocn_a_00151

https://doi.org/10.1162/jocn_a_00151

https://doi.org/10.1093/schbul/sbw036

https://doi.org/10.1093/schbul/sbw036

https://doi.org/10.1037/h0076780

https://doi.org/10.1037/xlm0000219


https://doi.org/10.1093/cercor/bhm195

https://doi.org/10.1016/j.neuropsychologia.2012.10.009

https://doi.org/10.1016/j.neuropsychologia.2012.10.009

https://doi.org/10.1093/Cercor/Bhm229

https://doi.org/10.1093/Cercor/Bhm229

https://doi.org/10.1093/cercor/5.2.123





https://doi.org/10.1017/s0140525x03000165

https://doi.org/10.1093/cercor/bhh050

https://doi.org/10.1093/cercor/bhh050

https://doi.org/10.1016/j.cub.2015.10.007





https://doi.org/10.1111/psyp.12437

Methods, 266, 166-170. https://doi.org/10.1016/j.jneumeth.2016.01.002

Tort, A. B., Komorowski, R., Eichenbaum, H., & Kopell, N. (2010).Measuring phase-amplitude coupling between neuronal oscillationsof different frequencies. Journal of Neurophysiology, 104(2), 1195-1210. https://doi.org/10.1152/jn.00106.2010

van Vugt, M. K., Schulze-Bonhage, A., Litt, B., Brandt, A., & Kahana,M. J. (2010). Hippocampal gamma oscillations increase with mem-ory load. The Journal of Neuroscience, 30(7), 2694-2699. https://doi.org/10.1523/JNEUROSCI.0567-09.2010

Vogel, E. K., & Machizawa, M. G. (2004). Neural activity predicts indi-vidual differences in visual working memory capacity. Nature,428(6984), 748-751. https://doi.org/10.1038/nature02447

Voytek, B., Canolty, R. T., Shestyuk, A., Crone, N. E., Parvizi, J., &Knight, R. T. (2010). Shifts in gamma phase-amplitude couplingfrequency from theta to alpha over posterior cortex during visualtasks. Frontiers in Human Neuroscience, 4, 191. https://doi.org/10.3389/fnhum.2010.00191

Winkler, I., Debener, S., Muller, K. R., & Tangermann, M. (2015). Onthe influence of high-pass filtering on ICA-based artifact reductionin EEG-ERP. Annual International Conference IEEE EngineringMedical Biology Social, 2015, 4101-4105. https://doi.org/10.1109/EMBC.2015.7319296

Yuval-Greenberg, S., & Deouell, L. Y. (2009). The broadband-transientinduced gamma-band response in scalp EEG reflects the executionof saccades. Brain Topography, 22(1), 3-6. https://doi.org/10.1007/s10548-009-0077-6

Yuval-Greenberg, S., Tomer, O., Keren, A. S., Nelken, I., & Deouell, L.Y. (2008). Transient induced gamma-band response in EEG as amanifestation of miniature saccades. Neuron, 58(3), 429-441.https://doi.org/10.1016/j.neuron.2008.03.027

Open Practices Statement

The data and materials for all experiments are available upon requestas well as on the NIMH National Data Archive (https://nda.nih.gov/).None of the experiments was preregistered.

Publisher’s note Springer Nature remains neutral with regard to jurisdic-tional claims in published maps and institutional affiliations.


https://doi.org/10.1016/j.jneumeth.2016.01.002

https://doi.org/10.1016/j.jneumeth.2016.01.002

https://doi.org/10.1152/jn.00106.2010



https://doi.org/10.1038/nature02447



https://doi.org/10.1109/EMBC.2015.7319296

https://doi.org/10.1109/EMBC.2015.7319296

https://doi.org/10.1007/s10548-009-0077-6

https://doi.org/10.1007/s10548-009-0077-6


http://creativecommons.org/licenses/by/4.0/

Documents

Neurocognitive subprocesses of working memory performance